WEBVTT 00:00:00.000 --> 00:00:07.952 align:middle line:84% [SQUEAKING] [RUSTLING] [CLICKING] 00:00:07.952 --> 00:00:18.400 align:middle line:90% 00:00:18.400 --> 00:00:20.470 align:middle line:84% BEN OLKEN: So just to remind you where we were, 00:00:20.470 --> 00:00:22.840 align:middle line:84% we were talking about the separation test. 00:00:22.840 --> 00:00:26.847 align:middle line:84% And we had gone through all the different-- 00:00:26.847 --> 00:00:28.930 align:middle line:84% all these different cases, testing for separation, 00:00:28.930 --> 00:00:29.860 align:middle line:90% or whatever. 00:00:29.860 --> 00:00:34.210 align:middle line:84% And the idea of testing for separation-- 00:00:34.210 --> 00:00:39.890 align:middle line:84% the idea was to say, if we have changes in household labor 00:00:39.890 --> 00:00:43.160 align:middle line:84% supply, does that affect the amount of labor 00:00:43.160 --> 00:00:46.160 align:middle line:90% demanded on the household farms? 00:00:46.160 --> 00:00:48.290 align:middle line:84% And then, we talked about some of the challenges 00:00:48.290 --> 00:00:50.248 align:middle line:84% with some of the empirics in the Benjamin paper 00:00:50.248 --> 00:00:51.720 align:middle line:84% that you guys read for last time. 00:00:51.720 --> 00:00:55.550 align:middle line:84% So the idea of this LeFave and Thomas paper is basically it 00:00:55.550 --> 00:00:58.070 align:middle line:84% just literally takes the model, et cetera, off the shelf 00:00:58.070 --> 00:01:03.990 align:middle line:84% from the Benjamin paper, but redoes it using panel data. 00:01:03.990 --> 00:01:07.670 align:middle line:84% And so why-- so I don't-- did we discuss this at all at the end 00:01:07.670 --> 00:01:08.450 align:middle line:90% of last class? 00:01:08.450 --> 00:01:10.110 align:middle line:90% Yes, no? 00:01:10.110 --> 00:01:13.600 align:middle line:84% Yes, we did, a little bit but not a lot. 00:01:13.600 --> 00:01:15.450 align:middle line:90% OK, so just to sum up then-- 00:01:15.450 --> 00:01:19.005 align:middle line:84% OK, so, Tara, so since you answered the question. 00:01:19.005 --> 00:01:22.212 align:middle line:84% So what was the idea of why was the panel data helping us? 00:01:22.212 --> 00:01:23.600 align:middle line:90% AUDIENCE: [INAUDIBLE] effects? 00:01:23.600 --> 00:01:24.217 align:middle line:90% BEN OLKEN: Sorry, Shawna. 00:01:24.217 --> 00:01:24.967 align:middle line:90% AUDIENCE: It's OK. 00:01:24.967 --> 00:01:30.827 align:middle line:84% We change [INAUDIBLE] effects for individuals or farms. 00:01:30.827 --> 00:01:32.660 align:middle line:84% BEN OLKEN: Exactly, and why is that helping? 00:01:32.660 --> 00:01:36.200 align:middle line:84% What is the empirical concern that that's helping? 00:01:36.200 --> 00:01:38.160 align:middle line:84% AUDIENCE: That there are unobserved 00:01:38.160 --> 00:01:41.710 align:middle line:84% individual characteristics of households. 00:01:41.710 --> 00:01:43.210 align:middle line:84% BEN OLKEN: Yeah, exactly, so there's 00:01:43.210 --> 00:01:44.470 align:middle line:90% unobservable characteristics. 00:01:44.470 --> 00:01:47.890 align:middle line:84% So maybe there were people who had really good land. 00:01:47.890 --> 00:01:49.307 align:middle line:84% Those people would have more kids. 00:01:49.307 --> 00:01:51.890 align:middle line:84% And they would also have-- we'd be sure they'd have more kids. 00:01:51.890 --> 00:01:53.350 align:middle line:84% They'd also have more labor demand. 00:01:53.350 --> 00:01:54.760 align:middle line:84% That's the unobserved characteristic 00:01:54.760 --> 00:01:56.135 align:middle line:84% that we'd be dealing with, right? 00:01:56.135 --> 00:01:57.760 align:middle line:84% So fixed effects basically allow you 00:01:57.760 --> 00:02:04.520 align:middle line:84% to deal with that because we can use variation over time. 00:02:04.520 --> 00:02:08.350 align:middle line:84% So the idea is that, basically, we're going to-- 00:02:08.350 --> 00:02:09.820 align:middle line:90% so in a very-- 00:02:09.820 --> 00:02:11.350 align:middle line:84% if we basically ran this regression 00:02:11.350 --> 00:02:13.600 align:middle line:84% but there was unobserved land quality that was correlated 00:02:13.600 --> 00:02:15.725 align:middle line:84% with both labor demand and people in the household, 00:02:15.725 --> 00:02:18.460 align:middle line:84% with fixed effects, we're going to include that. 00:02:18.460 --> 00:02:21.640 align:middle line:84% It's equivalent of essentially demeaning things. 00:02:21.640 --> 00:02:23.650 align:middle line:84% And so that concern is no longer there. 00:02:23.650 --> 00:02:28.470 align:middle line:90% 00:02:28.470 --> 00:02:31.260 align:middle line:84% So does this strike you as a more plausible 00:02:31.260 --> 00:02:32.130 align:middle line:90% empirical setting? 00:02:32.130 --> 00:02:33.930 align:middle line:90% Yes, no? 00:02:33.930 --> 00:02:35.430 align:middle line:84% What would you think about for this? 00:02:35.430 --> 00:02:44.530 align:middle line:90% 00:02:44.530 --> 00:02:45.910 align:middle line:90% Yes, no? 00:02:45.910 --> 00:02:47.560 align:middle line:84% Well, clearly, it's more plausible. 00:02:47.560 --> 00:02:50.360 align:middle line:90% Does it strike you as plausible? 00:02:50.360 --> 00:02:52.430 align:middle line:84% So the ideas we are going to use changes-- 00:02:52.430 --> 00:02:56.770 align:middle line:84% so first of all, what's identifying this, to be clear, 00:02:56.770 --> 00:03:00.760 align:middle line:84% is changes over time in household composition 00:03:00.760 --> 00:03:02.680 align:middle line:90% within a given household. 00:03:02.680 --> 00:03:06.850 align:middle line:84% So we have to have a panel structure where household 00:03:06.850 --> 00:03:08.960 align:middle line:90% composition is changing. 00:03:08.960 --> 00:03:12.310 align:middle line:84% And so the question is, if we get more or fewer people 00:03:12.310 --> 00:03:14.470 align:middle line:84% in our household, is that associated 00:03:14.470 --> 00:03:17.650 align:middle line:90% with changes in labor demand? 00:03:17.650 --> 00:03:19.870 align:middle line:84% So first of all, does some want to argue 00:03:19.870 --> 00:03:22.330 align:middle line:90% that that's not plausible? 00:03:22.330 --> 00:03:25.280 align:middle line:84% What would be a problem with that identification idea? 00:03:25.280 --> 00:03:26.030 align:middle line:90% Yeah, Whitney? 00:03:26.030 --> 00:03:28.490 align:middle line:84% AUDIENCE: I think we haven't completely fixed 00:03:28.490 --> 00:03:30.960 align:middle line:90% the endogenous family size-- 00:03:30.960 --> 00:03:32.270 align:middle line:90% BEN OLKEN: So the same arm. 00:03:32.270 --> 00:03:37.790 align:middle line:84% AUDIENCE: So if, for example, your household 00:03:37.790 --> 00:03:41.180 align:middle line:84% isn't doing well, you're like, my relative 00:03:41.180 --> 00:03:42.080 align:middle line:90% went out to go work. 00:03:42.080 --> 00:03:43.940 align:middle line:84% But now, I need more help or something. 00:03:43.940 --> 00:03:46.140 align:middle line:84% I'm going to ask them to come home. 00:03:46.140 --> 00:03:48.663 align:middle line:84% And so now, my household size has increased, 00:03:48.663 --> 00:03:49.580 align:middle line:90% but that was because-- 00:03:49.580 --> 00:03:51.740 align:middle line:84% BEN OLKEN: Exactly, so imagine I-- 00:03:51.740 --> 00:03:54.098 align:middle line:84% so just so everyone-- if everyone heard, so basically 00:03:54.098 --> 00:03:56.390 align:middle line:84% the idea is imagine I have this big labor-demand shock. 00:03:56.390 --> 00:03:57.370 align:middle line:90% I need more land. 00:03:57.370 --> 00:03:59.120 align:middle line:84% I invite some cousin to come stay with me. 00:03:59.120 --> 00:04:00.680 align:middle line:84% Now, my household size has increased. 00:04:00.680 --> 00:04:03.290 align:middle line:84% That's a violation of this, so that's exactly right. 00:04:03.290 --> 00:04:06.208 align:middle line:90% So what they're going to do-- 00:04:06.208 --> 00:04:07.500 align:middle line:90% so that's an endogenous change. 00:04:07.500 --> 00:04:08.990 align:middle line:90% That's what we don't want. 00:04:08.990 --> 00:04:11.180 align:middle line:90% What would be an-- 00:04:11.180 --> 00:04:12.140 align:middle line:90% so that's a problem. 00:04:12.140 --> 00:04:15.042 align:middle line:84% What would be exogenous change in household composition? 00:04:15.042 --> 00:04:18.140 align:middle line:90% 00:04:18.140 --> 00:04:18.649 align:middle line:90% Yeah? 00:04:18.649 --> 00:04:20.290 align:middle line:84% AUDIENCE: Someone who surprisingly dies. 00:04:20.290 --> 00:04:20.589 align:middle line:90% BEN OLKEN: What? 00:04:20.589 --> 00:04:20.920 align:middle line:90% Sorry? 00:04:20.920 --> 00:04:22.420 align:middle line:84% AUDIENCE: Someone surprisingly dies. 00:04:22.420 --> 00:04:24.220 align:middle line:84% BEN OLKEN: Yes, surprise deaths that 00:04:24.220 --> 00:04:26.530 align:middle line:84% would absolutely be a change in household composition. 00:04:26.530 --> 00:04:27.730 align:middle line:90% What else? 00:04:27.730 --> 00:04:30.355 align:middle line:84% What's a more common change in household composition? 00:04:30.355 --> 00:04:31.480 align:middle line:90% AUDIENCE: Unplanned births. 00:04:31.480 --> 00:04:32.147 align:middle line:90% BEN OLKEN: What? 00:04:32.147 --> 00:04:33.280 align:middle line:90% AUDIENCE: Unplanned births. 00:04:33.280 --> 00:04:35.400 align:middle line:84% BEN OLKEN: Yeah, unplanned births, but actually-- 00:04:35.400 --> 00:04:36.400 align:middle line:90% OK, so I agree with you. 00:04:36.400 --> 00:04:38.290 align:middle line:84% That's more common than surprised, 00:04:38.290 --> 00:04:40.040 align:middle line:84% although actually I'm not sure about that. 00:04:40.040 --> 00:04:42.530 align:middle line:84% I'm not sure which is more common. 00:04:42.530 --> 00:04:46.775 align:middle line:84% But is there something even more common than that they can use? 00:04:46.775 --> 00:04:47.275 align:middle line:90% Yeah? 00:04:47.275 --> 00:04:48.730 align:middle line:90% AUDIENCE: Gender of the kid. 00:04:48.730 --> 00:04:48.970 align:middle line:90% BEN OLKEN: What? 00:04:48.970 --> 00:04:49.470 align:middle line:90% Sorry? 00:04:49.470 --> 00:04:50.832 align:middle line:90% AUDIENCE: Gender of the kid. 00:04:50.832 --> 00:04:53.290 align:middle line:84% BEN OLKEN: Potentially, if you basically think that-- yeah. 00:04:53.290 --> 00:04:54.610 align:middle line:84% But I think that one thing you guys 00:04:54.610 --> 00:04:56.777 align:middle line:84% are forgetting is that not everyone in the household 00:04:56.777 --> 00:04:59.290 align:middle line:84% actually is of working age, and that's actually 00:04:59.290 --> 00:05:00.520 align:middle line:90% what they're going to use. 00:05:00.520 --> 00:05:02.500 align:middle line:84% So you guys are thinking of changing, 00:05:02.500 --> 00:05:04.090 align:middle line:84% people coming into the household and out of the household, 00:05:04.090 --> 00:05:04.540 align:middle line:90% and you're right. 00:05:04.540 --> 00:05:06.915 align:middle line:84% Those are the kinds of shocks that we would use for that. 00:05:06.915 --> 00:05:08.560 align:middle line:90% But you get a new-- 00:05:08.560 --> 00:05:13.452 align:middle line:84% an infant is not going to affect your household labor supply. 00:05:13.452 --> 00:05:15.160 align:middle line:84% Kids have to be a certain age before they 00:05:15.160 --> 00:05:17.770 align:middle line:84% can start contributing to labor on the farm. 00:05:17.770 --> 00:05:20.620 align:middle line:84% And likewise, at the other end, after people get a certain age, 00:05:20.620 --> 00:05:24.370 align:middle line:84% they're typically not supplying as much labor. 00:05:24.370 --> 00:05:26.890 align:middle line:84% So what they're going to do is basically 00:05:26.890 --> 00:05:30.700 align:middle line:84% they're going to use the aging of household members 00:05:30.700 --> 00:05:40.420 align:middle line:84% as surely predictable changes in your household labor supply, 00:05:40.420 --> 00:05:43.000 align:middle line:84% but changes that are uncorrelated 00:05:43.000 --> 00:05:45.450 align:middle line:90% with labor-demand shocks. 00:05:45.450 --> 00:05:47.020 align:middle line:84% So the idea is even if you thought 00:05:47.020 --> 00:05:48.745 align:middle line:84% like, oh, we need more kids in the farm, 00:05:48.745 --> 00:05:52.000 align:middle line:84% and we're going to have more kids, there's a time lag there. 00:05:52.000 --> 00:05:55.992 align:middle line:84% So we have a-- they have a kid who starts at age 0. 00:05:55.992 --> 00:05:57.700 align:middle line:84% They don't start working until they're, I 00:05:57.700 --> 00:05:59.770 align:middle line:84% don't know, say, teenagers or whatever. 00:05:59.770 --> 00:06:03.860 align:middle line:84% That's when the-- that shift from being little kids 00:06:03.860 --> 00:06:05.740 align:middle line:84% that can't work to teenagers that can work 00:06:05.740 --> 00:06:09.910 align:middle line:84% or adults that can work, that's the variation they want to use. 00:06:09.910 --> 00:06:12.160 align:middle line:90% Does that make sense? 00:06:12.160 --> 00:06:14.350 align:middle line:84% So that's what they're going to do. 00:06:14.350 --> 00:06:16.122 align:middle line:84% They also have just really terrific data. 00:06:16.122 --> 00:06:17.830 align:middle line:84% And one thing I will mention, by the way, 00:06:17.830 --> 00:06:20.460 align:middle line:90% is this was an example of-- 00:06:20.460 --> 00:06:23.965 align:middle line:84% as far as I know, and the authors correct me 00:06:23.965 --> 00:06:24.840 align:middle line:90% if I'm wrong on this. 00:06:24.840 --> 00:06:26.490 align:middle line:84% But I'm pretty sure this was data 00:06:26.490 --> 00:06:28.990 align:middle line:84% that was collected for a totally different purpose. 00:06:28.990 --> 00:06:30.865 align:middle line:84% This was data that was collected because they 00:06:30.865 --> 00:06:33.760 align:middle line:84% were doing a long run study of iron supplementation in Central 00:06:33.760 --> 00:06:34.260 align:middle line:90% Java. 00:06:34.260 --> 00:06:35.677 align:middle line:84% And so they wanted to look at what 00:06:35.677 --> 00:06:38.190 align:middle line:84% was the effects of iron supplementation on labor supply 00:06:38.190 --> 00:06:39.357 align:middle line:90% and a bunch of other things. 00:06:39.357 --> 00:06:41.148 align:middle line:84% And they had all this very rich [INAUDIBLE] 00:06:41.148 --> 00:06:42.900 align:middle line:84% data, and then realized they could use it 00:06:42.900 --> 00:06:44.150 align:middle line:90% to answer this other question. 00:06:44.150 --> 00:06:48.630 align:middle line:84% And so I actually think one useful thing for you 00:06:48.630 --> 00:06:51.090 align:middle line:84% to think about in terms of data sources in general 00:06:51.090 --> 00:06:53.130 align:middle line:84% is that there are all of these data sets 00:06:53.130 --> 00:06:58.450 align:middle line:84% that have been collected for all kinds of other things. 00:06:58.450 --> 00:07:00.200 align:middle line:84% Someone is doing a study of dust and such, 00:07:00.200 --> 00:07:03.350 align:middle line:84% and collects a bazillion variables to study it. 00:07:03.350 --> 00:07:05.690 align:middle line:84% And those data sets are often incredibly rich 00:07:05.690 --> 00:07:07.470 align:middle line:84% and can be used for other purposes. 00:07:07.470 --> 00:07:10.610 align:middle line:84% And so actually, if you look, for example, in the JPL 00:07:10.610 --> 00:07:12.110 align:middle line:84% Dataverse, for example, we have lots 00:07:12.110 --> 00:07:13.693 align:middle line:84% of different websites from JPL studies 00:07:13.693 --> 00:07:15.560 align:middle line:84% or other sites that compile these as well. 00:07:15.560 --> 00:07:20.157 align:middle line:84% And it's worth thinking about whether or not 00:07:20.157 --> 00:07:22.490 align:middle line:84% you have data sets that can be collected for one purpose 00:07:22.490 --> 00:07:24.300 align:middle line:90% and repurposed. 00:07:24.300 --> 00:07:26.215 align:middle line:84% I did this a bit in one of my own papers. 00:07:26.215 --> 00:07:27.590 align:middle line:84% I have a paper on social capital, 00:07:27.590 --> 00:07:29.870 align:middle line:84% for example, which was repurposing 00:07:29.870 --> 00:07:31.580 align:middle line:90% a data set that was used for-- 00:07:31.580 --> 00:07:35.450 align:middle line:84% that I collected for a study about corruption. 00:07:35.450 --> 00:07:39.620 align:middle line:84% So you can think about how these data sets can be repurposed. 00:07:39.620 --> 00:07:42.410 align:middle line:84% And I expect there's lots of good data out there 00:07:42.410 --> 00:07:43.705 align:middle line:90% that hasn't been exploited. 00:07:43.705 --> 00:07:45.080 align:middle line:84% And this is a really nice example 00:07:45.080 --> 00:07:47.360 align:middle line:84% of a paper that was ultimately published in Econometrica using 00:07:47.360 --> 00:07:49.568 align:middle line:84% a data set that was collected for some other purpose. 00:07:49.568 --> 00:07:51.920 align:middle line:90% 00:07:51.920 --> 00:07:53.350 align:middle line:90% So what's the regression? 00:07:53.350 --> 00:07:55.100 align:middle line:84% The regression is going to be labor demand 00:07:55.100 --> 00:07:57.380 align:middle line:84% on the farm as a function of number 00:07:57.380 --> 00:08:01.250 align:middle line:84% of people in the household of different size bins. 00:08:01.250 --> 00:08:06.470 align:middle line:84% And they're going to control for household fixed effects, 00:08:06.470 --> 00:08:09.500 align:middle line:84% and they'll also control for community times time fixed 00:08:09.500 --> 00:08:10.770 align:middle line:90% effects. 00:08:10.770 --> 00:08:13.810 align:middle line:84% So they're controlling for the fact that maybe-- 00:08:13.810 --> 00:08:15.260 align:middle line:84% so why would you care about this? 00:08:15.260 --> 00:08:16.060 align:middle line:84% If you imagine there were, I don't 00:08:16.060 --> 00:08:17.390 align:middle line:84% know, secular growth in household size, 00:08:17.390 --> 00:08:18.300 align:middle line:84% not that that's necessarily true, 00:08:18.300 --> 00:08:19.910 align:middle line:84% but if there was secular growth in household sizes 00:08:19.910 --> 00:08:21.290 align:middle line:84% and there was secular growth in labor demand, 00:08:21.290 --> 00:08:23.480 align:middle line:84% you'd want to control for time effects too. 00:08:23.480 --> 00:08:25.290 align:middle line:84% So that's going to control for that. 00:08:25.290 --> 00:08:26.840 align:middle line:84% And they're going to parameterize 00:08:26.840 --> 00:08:29.150 align:middle line:90% this N into different age bins. 00:08:29.150 --> 00:08:31.280 align:middle line:84% And to isolate these exogenous changes, 00:08:31.280 --> 00:08:32.780 align:middle line:84% they're going to restrict themselves 00:08:32.780 --> 00:08:35.683 align:middle line:84% to households with no change in household 00:08:35.683 --> 00:08:37.850 align:middle line:84% size due to membership because of migration, births, 00:08:37.850 --> 00:08:38.350 align:middle line:90% or deaths. 00:08:38.350 --> 00:08:41.360 align:middle line:84% They're going to just look at the changes 00:08:41.360 --> 00:08:45.140 align:middle line:90% due to the natural aging. 00:08:45.140 --> 00:08:47.810 align:middle line:90% Is that clear? 00:08:47.810 --> 00:08:53.153 align:middle line:84% All right, OK, so this is a funny regression 00:08:53.153 --> 00:08:55.070 align:middle line:84% because there's lots and lots of coefficients. 00:08:55.070 --> 00:08:56.653 align:middle line:84% Here are the different age bins like 0 00:08:56.653 --> 00:08:59.900 align:middle line:84% to 14, 15 to 19, 20 to 34, et cetera, et cetera. 00:08:59.900 --> 00:09:07.800 align:middle line:84% This is household farm demand on these various things. 00:09:07.800 --> 00:09:11.900 align:middle line:84% And in general, you see that farm demand 00:09:11.900 --> 00:09:14.240 align:middle line:90% is increasing as you-- 00:09:14.240 --> 00:09:17.192 align:middle line:90% 00:09:17.192 --> 00:09:20.420 align:middle line:84% log labor demand as a function of the number of people 00:09:20.420 --> 00:09:23.390 align:middle line:84% in these different groups is increasing 00:09:23.390 --> 00:09:26.630 align:middle line:84% for all the different age ranges, OK, fine. 00:09:26.630 --> 00:09:28.850 align:middle line:84% If you use the variation from aging only, 00:09:28.850 --> 00:09:31.228 align:middle line:84% it's-- the standard errors are a little bit smaller. 00:09:31.228 --> 00:09:33.020 align:middle line:84% By the way, they do this separately for men 00:09:33.020 --> 00:09:34.940 align:middle line:90% and for women. 00:09:34.940 --> 00:09:37.280 align:middle line:90% It does look like you get-- 00:09:37.280 --> 00:09:39.770 align:middle line:84% for example, in this age range the 15- to 19-year-olds, 00:09:39.770 --> 00:09:41.210 align:middle line:84% it looks like, for example, you're 00:09:41.210 --> 00:09:45.800 align:middle line:84% getting impacts on labor demand from boys but not from girls. 00:09:45.800 --> 00:09:48.860 align:middle line:84% In other age ranges, you get it from both. 00:09:48.860 --> 00:09:52.520 align:middle line:84% But the key thing actually of all of these joint tests. 00:09:52.520 --> 00:09:54.950 align:middle line:84% So the thing you want to do is reject-- 00:09:54.950 --> 00:09:59.360 align:middle line:84% if we want to reject separation, we just want to do a joint test 00:09:59.360 --> 00:10:00.230 align:middle line:90% that-- 00:10:00.230 --> 00:10:02.690 align:middle line:84% of all of these demographic variables, not predicting labor 00:10:02.690 --> 00:10:03.710 align:middle line:90% demand. 00:10:03.710 --> 00:10:05.853 align:middle line:84% And that's what this is right here. 00:10:05.853 --> 00:10:08.020 align:middle line:84% And so you can see they can strongly reject the null 00:10:08.020 --> 00:10:11.360 align:middle line:84% of separation that basically-- so on net, 00:10:11.360 --> 00:10:14.060 align:middle line:90% all those dummies are-- 00:10:14.060 --> 00:10:17.690 align:middle line:84% all those variables of the number of households 00:10:17.690 --> 00:10:20.750 align:middle line:90% of different sizes are-- 00:10:20.750 --> 00:10:25.165 align:middle line:90% 00:10:25.165 --> 00:10:26.540 align:middle line:84% they can reject the null of that. 00:10:26.540 --> 00:10:32.410 align:middle line:90% 00:10:32.410 --> 00:10:34.270 align:middle line:84% In some sense, this table of a bunch 00:10:34.270 --> 00:10:38.810 align:middle line:84% of p-values of joint tests is sometimes the main results. 00:10:38.810 --> 00:10:41.950 align:middle line:84% So substantively, it actually looks 00:10:41.950 --> 00:10:46.270 align:middle line:84% like we can reject separation with much better data 00:10:46.270 --> 00:10:48.310 align:middle line:90% and a better empirical strategy. 00:10:48.310 --> 00:10:49.300 align:middle line:90% Yeah, Kyle? 00:10:49.300 --> 00:10:50.740 align:middle line:84% AUDIENCE: Why do you think it published so well? 00:10:50.740 --> 00:10:51.970 align:middle line:84% Was it just like a really important-- 00:10:51.970 --> 00:10:52.455 align:middle line:90% BEN OLKEN: What? 00:10:52.455 --> 00:10:52.940 align:middle line:90% Sorry? 00:10:52.940 --> 00:10:55.440 align:middle line:84% AUDIENCE: Why do you think this paper was published so well? 00:10:55.440 --> 00:10:57.130 align:middle line:84% Is that a very important question? 00:10:57.130 --> 00:10:57.670 align:middle line:90% Or? 00:10:57.670 --> 00:10:58.587 align:middle line:90% BEN OLKEN: I think so. 00:10:58.587 --> 00:11:00.910 align:middle line:84% I think that we'd all been teaching that Benjamin 00:11:00.910 --> 00:11:01.990 align:middle line:90% paper for-- 00:11:01.990 --> 00:11:05.290 align:middle line:84% since 1982 and having to complain about the 00:11:05.290 --> 00:11:07.403 align:middle line:84% empirics not being up to modern standards. 00:11:07.403 --> 00:11:09.070 align:middle line:84% And so having someone who can say, look, 00:11:09.070 --> 00:11:13.420 align:middle line:84% this is a classic development economics question, 00:11:13.420 --> 00:11:15.797 align:middle line:84% and I'm just going to nail the empirical of that, 00:11:15.797 --> 00:11:16.630 align:middle line:90% that's a nice paper. 00:11:16.630 --> 00:11:19.490 align:middle line:90% 00:11:19.490 --> 00:11:20.990 align:middle line:84% And actually, that could be like-- 00:11:20.990 --> 00:11:26.190 align:middle line:84% if you can find great, classic questions with-- 00:11:26.190 --> 00:11:28.370 align:middle line:84% that are waiting for a good empirical test, 00:11:28.370 --> 00:11:32.633 align:middle line:84% I think that's a great result. I would have totally been 00:11:32.633 --> 00:11:33.800 align:middle line:90% happy to publish this paper. 00:11:33.800 --> 00:11:34.592 align:middle line:90% It's a great paper. 00:11:34.592 --> 00:11:38.480 align:middle line:90% 00:11:38.480 --> 00:11:40.380 align:middle line:90% Other questions? 00:11:40.380 --> 00:11:41.880 align:middle line:84% And I was so excited to finally be 00:11:41.880 --> 00:11:44.338 align:middle line:84% able to update my slides when it came out and be like, aha, 00:11:44.338 --> 00:11:46.507 align:middle line:84% we can now actually say something a little more 00:11:46.507 --> 00:11:47.340 align:middle line:90% positive about that. 00:11:47.340 --> 00:11:50.500 align:middle line:90% And so I think that's great. 00:11:50.500 --> 00:11:55.120 align:middle line:84% And you'll see lots of papers that have that feel of, 00:11:55.120 --> 00:11:57.333 align:middle line:84% here's a classic theoretical test, 00:11:57.333 --> 00:11:59.500 align:middle line:84% and then someone's going to really nail the empiric. 00:11:59.500 --> 00:12:03.700 align:middle line:84% For example, if you think about the poverty trap section, 00:12:03.700 --> 00:12:05.650 align:middle line:84% I think Esther taught the Balboni et al. 00:12:05.650 --> 00:12:09.410 align:middle line:84% paper on the estimation of the poverty trap. 00:12:09.410 --> 00:12:11.012 align:middle line:84% And that was-- in some sense, what's 00:12:11.012 --> 00:12:12.220 align:middle line:90% so exciting about that paper? 00:12:12.220 --> 00:12:14.890 align:middle line:84% Well, we all knew that we could write-- 00:12:14.890 --> 00:12:20.920 align:middle line:84% we could all right, do this graph of KT plus 1 versus KT, 00:12:20.920 --> 00:12:23.680 align:middle line:84% and drew something kind of funky that looked like that. 00:12:23.680 --> 00:12:27.390 align:middle line:84% But actually really nailing kind of what the shape of that curve 00:12:27.390 --> 00:12:29.140 align:middle line:84% looks like and establishing whether or not 00:12:29.140 --> 00:12:30.150 align:middle line:84% there are multiple equilibria or not 00:12:30.150 --> 00:12:30.940 align:middle line:84% is something that I think everyone 00:12:30.940 --> 00:12:31.940 align:middle line:90% is really excited about. 00:12:31.940 --> 00:12:33.612 align:middle line:84% So I think that's another aspect, that's 00:12:33.612 --> 00:12:35.320 align:middle line:84% another reason that paper is so exciting, 00:12:35.320 --> 00:12:38.140 align:middle line:84% is because that basic sort of theoretical framework we've 00:12:38.140 --> 00:12:40.060 align:middle line:84% had for a long time, but actually 00:12:40.060 --> 00:12:41.710 align:middle line:84% being able to show it empirically 00:12:41.710 --> 00:12:43.310 align:middle line:84% in a convincing way is really nice. 00:12:43.310 --> 00:12:45.102 align:middle line:84% AUDIENCE: It's kind of a significant amount 00:12:45.102 --> 00:12:45.970 align:middle line:90% or something? 00:12:45.970 --> 00:12:46.887 align:middle line:90% BEN OLKEN: I think so. 00:12:46.887 --> 00:12:47.920 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:12:47.920 --> 00:12:49.000 align:middle line:84% BEN OLKEN: On both those examples. 00:12:49.000 --> 00:12:49.800 align:middle line:90% AUDIENCE: Yeah. 00:12:49.800 --> 00:12:50.467 align:middle line:90% BEN OLKEN: Yeah. 00:12:50.467 --> 00:12:52.770 align:middle line:90% 00:12:52.770 --> 00:12:58.270 align:middle line:84% But again, the key there is finding a classic-- 00:12:58.270 --> 00:12:59.830 align:middle line:84% it's not some random theory paper 00:12:59.830 --> 00:13:01.510 align:middle line:90% that no one's ever heard of. 00:13:01.510 --> 00:13:03.978 align:middle line:84% This is a classic poverty trap model that everybody knows. 00:13:03.978 --> 00:13:05.770 align:middle line:84% This one, this Benjamin paper, is something 00:13:05.770 --> 00:13:08.832 align:middle line:84% that we've all been teaching as one 00:13:08.832 --> 00:13:10.540 align:middle line:84% of these key questions in labor economics 00:13:10.540 --> 00:13:11.873 align:middle line:90% and development for a long time. 00:13:11.873 --> 00:13:13.627 align:middle line:84% So coming in and saying, this is going 00:13:13.627 --> 00:13:15.460 align:middle line:84% to be the definitive kind of empirical piece 00:13:15.460 --> 00:13:16.600 align:middle line:84% that goes along with this theoretical piece? 00:13:16.600 --> 00:13:17.751 align:middle line:90% That's awesome. 00:13:17.751 --> 00:13:20.340 align:middle line:90% 00:13:20.340 --> 00:13:23.080 align:middle line:90% Other questions? 00:13:23.080 --> 00:13:31.940 align:middle line:84% OK, all right, so what's going on? 00:13:31.940 --> 00:13:34.210 align:middle line:84% So if we buy that there's already separation failures, 00:13:34.210 --> 00:13:35.950 align:middle line:84% what is going on in these labor markets? 00:13:35.950 --> 00:13:36.900 align:middle line:90% Why is this happening? 00:13:36.900 --> 00:13:38.560 align:middle line:84% Well, if you think about the theory, 00:13:38.560 --> 00:13:41.170 align:middle line:84% it said something is going on that's making these labor 00:13:41.170 --> 00:13:42.400 align:middle line:90% markets not clear. 00:13:42.400 --> 00:13:44.090 align:middle line:90% That was kind of the key. 00:13:44.090 --> 00:13:47.860 align:middle line:84% That's why we got these separation failures. 00:13:47.860 --> 00:13:49.510 align:middle line:90% And so why? 00:13:49.510 --> 00:13:51.490 align:middle line:84% And so there was a bunch of papers 00:13:51.490 --> 00:13:54.790 align:middle line:84% that have been starting to understand and try to unpack 00:13:54.790 --> 00:13:57.560 align:middle line:84% in a little more detail, what is going on in these rural labor 00:13:57.560 --> 00:13:58.060 align:middle line:90% markets? 00:13:58.060 --> 00:13:59.590 align:middle line:90% Why are they not clearing? 00:13:59.590 --> 00:14:01.550 align:middle line:84% So I'm going to talk about a few of them. 00:14:01.550 --> 00:14:05.390 align:middle line:84% So in particular, I'm going to talk 00:14:05.390 --> 00:14:08.190 align:middle line:84% about several papers that look at this idea of nominal wage 00:14:08.190 --> 00:14:08.690 align:middle line:90% stickiness. 00:14:08.690 --> 00:14:10.773 align:middle line:84% So how many of you guys have heard of nominal wage 00:14:10.773 --> 00:14:13.200 align:middle line:90% stickiness, like for macro? 00:14:13.200 --> 00:14:13.830 align:middle line:90% Some of you? 00:14:13.830 --> 00:14:15.140 align:middle line:90% Not everyone. 00:14:15.140 --> 00:14:17.318 align:middle line:84% OK, so the basic-- so somebody's heard of it? 00:14:17.318 --> 00:14:18.360 align:middle line:90% You want to summarize it? 00:14:18.360 --> 00:14:20.052 align:middle line:90% What's the idea? 00:14:20.052 --> 00:14:23.180 align:middle line:84% AUDIENCE: You can't adjust wages downwards, or the nominal wage 00:14:23.180 --> 00:14:25.463 align:middle line:90% downwards, [INAUDIBLE]. 00:14:25.463 --> 00:14:27.130 align:middle line:84% BEN OLKEN: Because people don't like it. 00:14:27.130 --> 00:14:27.775 align:middle line:90% Right, exactly. 00:14:27.775 --> 00:14:29.650 align:middle line:84% So this is the idea that there's an asymmetry 00:14:29.650 --> 00:14:30.730 align:middle line:90% in wage adjustments. 00:14:30.730 --> 00:14:33.370 align:middle line:84% Everyone is happy to have their nominal wage go up. 00:14:33.370 --> 00:14:35.770 align:middle line:84% People do not like their nominal wage to go down. 00:14:35.770 --> 00:14:42.520 align:middle line:84% And for some reason, this is true about nominal wages 00:14:42.520 --> 00:14:44.440 align:middle line:84% and not so much about real wages. 00:14:44.440 --> 00:14:45.940 align:middle line:84% Because your nominal wage-- and even 00:14:45.940 --> 00:14:49.000 align:middle line:84% though it's harder for our models to think about, 00:14:49.000 --> 00:14:50.830 align:middle line:84% it is much easier for actual people 00:14:50.830 --> 00:14:53.262 align:middle line:84% to think about nominal numbers than real numbers, 00:14:53.262 --> 00:14:55.720 align:middle line:84% because your nominal wage is how much you're actually being 00:14:55.720 --> 00:14:57.042 align:middle line:90% paid every period or whatever. 00:14:57.042 --> 00:14:59.500 align:middle line:84% And that's really important for-- that has really important 00:14:59.500 --> 00:15:03.890 align:middle line:84% macro implications, as you may have learned about in macro, 00:15:03.890 --> 00:15:06.170 align:middle line:84% because it means that wages don't-- 00:15:06.170 --> 00:15:08.952 align:middle line:84% if there are shocks to labor markets, 00:15:08.952 --> 00:15:10.410 align:middle line:84% and you have nominal wage rigidity, 00:15:10.410 --> 00:15:12.050 align:middle line:84% and you have no inflation, then, basically, you 00:15:12.050 --> 00:15:13.190 align:middle line:84% have all these labor markets that 00:15:13.190 --> 00:15:15.690 align:middle line:84% are not clearing because you have people whose wages are not 00:15:15.690 --> 00:15:17.100 align:middle line:90% ticking downwards. 00:15:17.100 --> 00:15:20.355 align:middle line:84% And so at the macro level, that's often one of the main 00:15:20.355 --> 00:15:22.730 align:middle line:84% arguments why we should have a non-zero optimal inflation 00:15:22.730 --> 00:15:25.590 align:middle line:84% target, because we want to have a little bit of inflation 00:15:25.590 --> 00:15:27.350 align:middle line:84% so that these labor markets can-- 00:15:27.350 --> 00:15:29.600 align:middle line:84% because if we have nominal stickiness but not 00:15:29.600 --> 00:15:31.460 align:middle line:84% real stickiness, a little bit of inflation 00:15:31.460 --> 00:15:33.460 align:middle line:84% means the real wage can move around a little bit 00:15:33.460 --> 00:15:35.520 align:middle line:90% and those markets can clear. 00:15:35.520 --> 00:15:39.610 align:middle line:90% OK, so you might-- 00:15:39.610 --> 00:15:42.550 align:middle line:90% 00:15:42.550 --> 00:15:47.760 align:middle line:84% a natural question is, these ideas about nominal wage 00:15:47.760 --> 00:15:52.020 align:middle line:84% stickiness are very intuitive in very formal settings 00:15:52.020 --> 00:15:58.890 align:middle line:84% with formal wage contracts, and my wage is set-- 00:15:58.890 --> 00:16:00.840 align:middle line:90% MIT has set my wage. 00:16:00.840 --> 00:16:02.610 align:middle line:84% I have a letter saying what my salary is. 00:16:02.610 --> 00:16:06.240 align:middle line:90% It's all very by the book. 00:16:06.240 --> 00:16:09.090 align:middle line:84% It's even more unsurprising in cases 00:16:09.090 --> 00:16:11.790 align:middle line:84% where there are labor unions involved. 00:16:11.790 --> 00:16:15.270 align:middle line:84% But it's not obvious that this would translate into spot labor 00:16:15.270 --> 00:16:17.940 align:middle line:84% markets for agricultural labor in a developing country 00:16:17.940 --> 00:16:18.660 align:middle line:90% setting. 00:16:18.660 --> 00:16:20.340 align:middle line:84% It's like, we need to go hire some-- 00:16:20.340 --> 00:16:22.260 align:middle line:84% we need to go to the village center 00:16:22.260 --> 00:16:27.360 align:middle line:84% and pick up a few laborers to help clear our field today. 00:16:27.360 --> 00:16:29.790 align:middle line:84% Is that a setting where we expect nominal wage stickiness 00:16:29.790 --> 00:16:30.655 align:middle line:90% or not? 00:16:30.655 --> 00:16:31.530 align:middle line:90% And it's not obvious. 00:16:31.530 --> 00:16:35.360 align:middle line:90% 00:16:35.360 --> 00:16:37.860 align:middle line:84% But you can see why these would lead to separation failures, 00:16:37.860 --> 00:16:37.980 align:middle line:90% right? 00:16:37.980 --> 00:16:39.570 align:middle line:90% That was actually one of the-- 00:16:39.570 --> 00:16:40.224 align:middle line:90% so yes or no? 00:16:40.224 --> 00:16:42.016 align:middle line:84% Why would this lead to separation failures? 00:16:42.016 --> 00:16:47.610 align:middle line:90% 00:16:47.610 --> 00:16:48.903 align:middle line:90% OK, [INAUDIBLE]? 00:16:48.903 --> 00:16:50.820 align:middle line:84% AUDIENCE: If you're speaking of the household, 00:16:50.820 --> 00:16:53.010 align:middle line:84% you don't have a formal kind of [INAUDIBLE].. 00:16:53.010 --> 00:16:56.900 align:middle line:84% Within the household, you're not setting informal contracts. 00:16:56.900 --> 00:16:59.000 align:middle line:84% BEN OLKEN: Right, but more generally, 00:16:59.000 --> 00:17:02.090 align:middle line:84% which of those like various little graphs 00:17:02.090 --> 00:17:08.050 align:middle line:84% with the separation failures does this correspond to? 00:17:08.050 --> 00:17:10.005 align:middle line:90% Someone? 00:17:10.005 --> 00:17:11.579 align:middle line:90% Someone take a stab at this? 00:17:11.579 --> 00:17:15.730 align:middle line:90% 00:17:15.730 --> 00:17:17.440 align:middle line:84% AUDIENCE: The market wage is too high. 00:17:17.440 --> 00:17:19.982 align:middle line:84% BEN OLKEN: Yeah, the market wage is too high, exactly, right? 00:17:19.982 --> 00:17:22.812 align:middle line:84% So we'd like the wage to come down to clear, 00:17:22.812 --> 00:17:25.270 align:middle line:84% but it's stuck too high because of nominal wage stickiness. 00:17:25.270 --> 00:17:27.645 align:middle line:84% So it's the case where the nominal wage, the market wage, 00:17:27.645 --> 00:17:28.840 align:middle line:90% is too high. 00:17:28.840 --> 00:17:31.720 align:middle line:84% So we can't get enough-- we'd like more work. 00:17:31.720 --> 00:17:35.030 align:middle line:84% The households would like more work at the market wage, 00:17:35.030 --> 00:17:38.150 align:middle line:90% and they can't get it. 00:17:38.150 --> 00:17:42.330 align:middle line:84% All right, so [INAUDIBLE] has a paper 00:17:42.330 --> 00:17:44.490 align:middle line:84% testing this idea of whether there 00:17:44.490 --> 00:17:50.530 align:middle line:84% is nominal wage stickiness in rural labor markets. 00:17:50.530 --> 00:17:51.930 align:middle line:90% And here's her idea. 00:17:51.930 --> 00:17:54.030 align:middle line:90% Her idea is very simple. 00:17:54.030 --> 00:18:03.080 align:middle line:84% If you have nominal downward stickiness, then-- 00:18:03.080 --> 00:18:04.040 align:middle line:90% sorry, back up. 00:18:04.040 --> 00:18:07.250 align:middle line:84% She's going to use rainfall shocks 00:18:07.250 --> 00:18:11.010 align:middle line:90% as shocks to labor demand. 00:18:11.010 --> 00:18:13.953 align:middle line:84% So when you have good rainfall, we need lots of labor to clear, 00:18:13.953 --> 00:18:15.370 align:middle line:84% do all the agriculture activities. 00:18:15.370 --> 00:18:16.310 align:middle line:84% When we have bad rainfall, there's 00:18:16.310 --> 00:18:18.810 align:middle line:84% not that much to do because the crops didn't grow that well, 00:18:18.810 --> 00:18:21.000 align:middle line:90% so we have less labor demand. 00:18:21.000 --> 00:18:25.190 align:middle line:84% So we have rainfall shocks as a measure of labor demand. 00:18:25.190 --> 00:18:28.260 align:middle line:90% 00:18:28.260 --> 00:18:31.650 align:middle line:84% But with nominal downward stickiness, 00:18:31.650 --> 00:18:34.360 align:middle line:84% the sequence of shocks is going to matter. 00:18:34.360 --> 00:18:38.100 align:middle line:84% So in particular, through a positive shock and then 00:18:38.100 --> 00:18:40.500 align:middle line:84% a negative shock, the wages are going 00:18:40.500 --> 00:18:41.917 align:middle line:84% to go up after the positive shock, 00:18:41.917 --> 00:18:43.542 align:middle line:84% and they're going to be stuck too high. 00:18:43.542 --> 00:18:45.960 align:middle line:84% And they're not going to fall during the negative shock. 00:18:45.960 --> 00:18:48.690 align:middle line:84% Whereas if I have negative chocolate and a positive shock, 00:18:48.690 --> 00:18:49.650 align:middle line:90% I have a problem. 00:18:49.650 --> 00:18:52.192 align:middle line:84% And in particular, the key thing that you're going to look at 00:18:52.192 --> 00:18:56.670 align:middle line:84% is, in some sense, if I have a negative shock now-- 00:18:56.670 --> 00:18:59.190 align:middle line:90% so I want a low wage now-- 00:18:59.190 --> 00:19:02.430 align:middle line:84% then the lagged shock is going to matter. 00:19:02.430 --> 00:19:04.990 align:middle line:90% 00:19:04.990 --> 00:19:08.260 align:middle line:84% So if I want a low wage today because I got a negative shock, 00:19:08.260 --> 00:19:11.120 align:middle line:84% and I had a low wage yesterday, then we're good. 00:19:11.120 --> 00:19:13.150 align:middle line:84% But if I have a negative shock now, 00:19:13.150 --> 00:19:14.590 align:middle line:84% so the optimal wage is low today, 00:19:14.590 --> 00:19:17.230 align:middle line:84% but I had a positive shock in the past, 00:19:17.230 --> 00:19:21.200 align:middle line:84% then the wage is going to be artificially too high. 00:19:21.200 --> 00:19:23.950 align:middle line:90% Is that clear? 00:19:23.950 --> 00:19:25.480 align:middle line:90% Yes, no? 00:19:25.480 --> 00:19:26.797 align:middle line:90% [INAUDIBLE], no? 00:19:26.797 --> 00:19:28.630 align:middle line:84% AUDIENCE: But what if-- if it's sticky, then 00:19:28.630 --> 00:19:33.782 align:middle line:84% why would the first negative shock push the price down? 00:19:33.782 --> 00:19:34.490 align:middle line:90% What do you mean. 00:19:34.490 --> 00:19:37.670 align:middle line:84% Let's say you have two negative shots. 00:19:37.670 --> 00:19:39.620 align:middle line:84% Won't the first one be sticky as well? 00:19:39.620 --> 00:19:41.745 align:middle line:84% [INAUDIBLE] it's not going to fit. 00:19:41.745 --> 00:19:42.870 align:middle line:90% BEN OLKEN: It only sticks-- 00:19:42.870 --> 00:19:46.770 align:middle line:84% sticks the thing isn't-- so but it didn't go up. 00:19:46.770 --> 00:19:48.990 align:middle line:84% It's almost like the key of a negative, , 00:19:48.990 --> 00:19:50.227 align:middle line:90% it wasn't a positive shock. 00:19:50.227 --> 00:19:51.560 align:middle line:90% That's the thing to think about. 00:19:51.560 --> 00:19:55.290 align:middle line:84% So imagine we're just at some base case, 00:19:55.290 --> 00:20:00.570 align:middle line:84% and then, in period 1, we get just a neutral 00:20:00.570 --> 00:20:03.450 align:middle line:84% or a negative shock, so there's no increase in the wage. 00:20:03.450 --> 00:20:04.590 align:middle line:90% So the wage is over here. 00:20:04.590 --> 00:20:06.548 align:middle line:84% And in the other case, we get a positive shock, 00:20:06.548 --> 00:20:07.890 align:middle line:90% so the wage is over here. 00:20:07.890 --> 00:20:09.390 align:middle line:84% Now, the next question is, if we get 00:20:09.390 --> 00:20:12.810 align:middle line:84% a negative shock in the next period, over here, 00:20:12.810 --> 00:20:14.760 align:middle line:84% it'll still be the wage will be down here. 00:20:14.760 --> 00:20:16.990 align:middle line:84% Over here, the wage will be too high. 00:20:16.990 --> 00:20:19.302 align:middle line:84% So the key point is kind of that if you 00:20:19.302 --> 00:20:20.760 align:middle line:84% get a negative-- whereas, if you've 00:20:20.760 --> 00:20:23.052 align:middle line:84% got a positive shock in the next period, then this way, 00:20:23.052 --> 00:20:25.710 align:middle line:90% you can just flip right back up. 00:20:25.710 --> 00:20:28.410 align:middle line:84% So with positive shocks in the second period, 00:20:28.410 --> 00:20:29.940 align:middle line:90% the lag doesn't really matter. 00:20:29.940 --> 00:20:32.090 align:middle line:84% With negative shocks in the second period, 00:20:32.090 --> 00:20:33.840 align:middle line:84% then you have this persistence of the lag. 00:20:33.840 --> 00:20:34.710 align:middle line:84% But basically, the positive shock 00:20:34.710 --> 00:20:36.002 align:middle line:90% means the wage is way too high. 00:20:36.002 --> 00:20:38.780 align:middle line:84% That's the key asymmetry in our test. 00:20:38.780 --> 00:20:40.790 align:middle line:90% Does that make sense? 00:20:40.790 --> 00:20:42.810 align:middle line:84% OK, so if she's going to do this, 00:20:42.810 --> 00:20:45.127 align:middle line:84% she has data on a number of years from Indian districts 00:20:45.127 --> 00:20:46.710 align:middle line:84% where she has rainfall shocks, and she 00:20:46.710 --> 00:20:48.750 align:middle line:90% has the agricultural wage. 00:20:48.750 --> 00:20:51.090 align:middle line:90% And so this is kind of-- 00:20:51.090 --> 00:20:53.430 align:middle line:84% she goes has two-- this is from 1956 to '87, 00:20:53.430 --> 00:20:55.620 align:middle line:90% so she's got a lot of years. 00:20:55.620 --> 00:20:58.560 align:middle line:84% And this other, I guess, this from the World Bank data 00:20:58.560 --> 00:21:03.390 align:middle line:84% sets, and also this NSF data from '82 to 2008. 00:21:03.390 --> 00:21:06.630 align:middle line:84% And so the key thing is she's going 00:21:06.630 --> 00:21:10.437 align:middle line:84% to look at shock T minus 1 and shock at time T. 00:21:10.437 --> 00:21:12.520 align:middle line:84% So what are the key things we can learn from this? 00:21:12.520 --> 00:21:22.130 align:middle line:84% So first point is, positive shocks in the second period 00:21:22.130 --> 00:21:24.830 align:middle line:90% increase the wage. 00:21:24.830 --> 00:21:26.512 align:middle line:90% That's not surprising. 00:21:26.512 --> 00:21:32.370 align:middle line:84% It doesn't really seem to matter what the wage was last period. 00:21:32.370 --> 00:21:33.740 align:middle line:90% So if the shock is high-- 00:21:33.740 --> 00:21:37.350 align:middle line:84% this is all relative to 0, 0, like neutral-neutral. 00:21:37.350 --> 00:21:40.200 align:middle line:84% So if the wage was high this period, 00:21:40.200 --> 00:21:42.260 align:middle line:84% it doesn't matter if we were in a drought 00:21:42.260 --> 00:21:43.940 align:middle line:90% or had a high wage last period. 00:21:43.940 --> 00:21:46.675 align:middle line:90% Those are about the same. 00:21:46.675 --> 00:21:49.050 align:middle line:84% If anything, actually, it looks a little weird over here. 00:21:49.050 --> 00:21:49.740 align:middle line:90% It's negative. 00:21:49.740 --> 00:21:52.560 align:middle line:84% But certainly, it's not the case that drought last period 00:21:52.560 --> 00:21:54.510 align:middle line:90% is pulling the wage down today. 00:21:54.510 --> 00:21:57.990 align:middle line:90% 00:21:57.990 --> 00:22:04.410 align:middle line:84% On the other hand, if we have a drought this period, 00:22:04.410 --> 00:22:08.050 align:middle line:84% or a negative shock this period, the wage 00:22:08.050 --> 00:22:11.022 align:middle line:84% is higher if we had a positive shock last period. 00:22:11.022 --> 00:22:12.730 align:middle line:84% So that's where you can see, for example, 00:22:12.730 --> 00:22:16.900 align:middle line:84% like these coefficients, to the ones in this bin 00:22:16.900 --> 00:22:21.058 align:middle line:84% compared to the ones in this bin. 00:22:21.058 --> 00:22:24.923 align:middle line:90% OK, is that clear? 00:22:24.923 --> 00:22:26.340 align:middle line:84% I know this a little small, but it 00:22:26.340 --> 00:22:28.200 align:middle line:84% seems like guys have all your printouts, or your iPads, 00:22:28.200 --> 00:22:28.740 align:middle line:90% or whatever. 00:22:28.740 --> 00:22:30.870 align:middle line:84% So hopefully, you can read it, right? 00:22:30.870 --> 00:22:34.200 align:middle line:84% So that's-- this is the key asymmetry. 00:22:34.200 --> 00:22:36.040 align:middle line:84% Is like over here, it doesn't matter, 00:22:36.040 --> 00:22:38.420 align:middle line:84% but down here, it seems to matter. 00:22:38.420 --> 00:22:43.680 align:middle line:84% So the wage is stuck too high in the previous period. 00:22:43.680 --> 00:22:50.430 align:middle line:84% If you take this idea that this is about nominal rigidity, 00:22:50.430 --> 00:22:52.290 align:middle line:84% then these effects should be smaller 00:22:52.290 --> 00:22:53.730 align:middle line:90% if there's inflation going on. 00:22:53.730 --> 00:22:56.630 align:middle line:90% 00:22:56.630 --> 00:22:59.983 align:middle line:84% So this is a period over a 50 or, whatever, 30-year period. 00:22:59.983 --> 00:23:01.400 align:middle line:84% There's different inflation rates, 00:23:01.400 --> 00:23:05.500 align:middle line:84% so you can interact this whole thing with inflation. 00:23:05.500 --> 00:23:08.400 align:middle line:84% And if you remember, the key place 00:23:08.400 --> 00:23:10.080 align:middle line:84% we thought there was nominal stickiness 00:23:10.080 --> 00:23:13.770 align:middle line:84% was like negative shock this period, 00:23:13.770 --> 00:23:15.880 align:middle line:84% but positive shock in the last period. 00:23:15.880 --> 00:23:19.590 align:middle line:84% So you can interact all these shock things with an inflation 00:23:19.590 --> 00:23:21.630 align:middle line:84% measure, and the key thing you're going to find 00:23:21.630 --> 00:23:24.660 align:middle line:84% is a negative coefficient over here, which is basically 00:23:24.660 --> 00:23:28.530 align:middle line:84% that this effect of the wage being too high 00:23:28.530 --> 00:23:30.270 align:middle line:84% goes away when there's high inflation. 00:23:30.270 --> 00:23:38.570 align:middle line:90% 00:23:38.570 --> 00:23:41.377 align:middle line:84% OK, and the final question-- and I 00:23:41.377 --> 00:23:43.210 align:middle line:84% don't remember why they have acres per adult 00:23:43.210 --> 00:23:44.105 align:middle line:90% in the household, actually. 00:23:44.105 --> 00:23:45.070 align:middle line:90% I don't remember that. 00:23:45.070 --> 00:23:49.030 align:middle line:90% 00:23:49.030 --> 00:23:52.180 align:middle line:84% But the final point is, does this actually 00:23:52.180 --> 00:23:55.700 align:middle line:84% lead to changes in aggregate employment? 00:23:55.700 --> 00:23:57.880 align:middle line:84% So they have a measure of aggregate employment, 00:23:57.880 --> 00:24:01.300 align:middle line:84% and if you find that the shock is-- 00:24:01.300 --> 00:24:08.370 align:middle line:90% 00:24:08.370 --> 00:24:09.460 align:middle line:90% OK, so a couple of things. 00:24:09.460 --> 00:24:15.188 align:middle line:84% So something, actually, is going-- something 00:24:15.188 --> 00:24:16.480 align:middle line:90% is weird with this coefficient. 00:24:16.480 --> 00:24:18.130 align:middle line:84% It's negative over here and positive over 00:24:18.130 --> 00:24:19.547 align:middle line:84% here, so maybe there's something-- 00:24:19.547 --> 00:24:22.690 align:middle line:84% I suspect there's something off on that one over here. 00:24:22.690 --> 00:24:26.030 align:middle line:84% I expect that's probably negative. 00:24:26.030 --> 00:24:29.720 align:middle line:84% Yeah, so in general, positive shocks lead to more employment. 00:24:29.720 --> 00:24:31.790 align:middle line:84% Negative shocks lead to less employment. 00:24:31.790 --> 00:24:41.180 align:middle line:84% But when you have a negative shock today, a shock equals-- 00:24:41.180 --> 00:24:43.610 align:middle line:90% 00:24:43.610 --> 00:24:45.860 align:middle line:84% this table is a little funky because sometimes it's 0, 00:24:45.860 --> 00:24:47.027 align:middle line:90% and sometimes it's negative. 00:24:47.027 --> 00:24:49.670 align:middle line:84% I don't exactly know why she does that, frankly. 00:24:49.670 --> 00:24:54.710 align:middle line:84% But in general, when you have positive shocks 00:24:54.710 --> 00:25:02.210 align:middle line:84% in the past period and negative shocks 00:25:02.210 --> 00:25:05.930 align:middle line:84% today, you get more negative effects on employment 00:25:05.930 --> 00:25:10.730 align:middle line:84% than if you just had negative shocks today. 00:25:10.730 --> 00:25:17.360 align:middle line:84% So that lagged positive shock is giving you an extra bump in-- 00:25:17.360 --> 00:25:19.820 align:middle line:84% an additional negative employment effect 00:25:19.820 --> 00:25:26.690 align:middle line:84% above and beyond just the negative demand shock 00:25:26.690 --> 00:25:31.600 align:middle line:84% today, and that's because the price is stuck to high, right? 00:25:31.600 --> 00:25:34.280 align:middle line:84% And so what's going on there, what 00:25:34.280 --> 00:25:35.405 align:middle line:90% you'd like to have happen-- 00:25:35.405 --> 00:25:42.070 align:middle line:90% 00:25:42.070 --> 00:25:43.780 align:middle line:84% so what you'd like to have happen 00:25:43.780 --> 00:25:47.380 align:middle line:84% if you have a negative labor demand shock, 00:25:47.380 --> 00:25:51.550 align:middle line:84% the wage should fall, and employment should also fall. 00:25:51.550 --> 00:25:56.260 align:middle line:84% But if the wage is stuck too high, 00:25:56.260 --> 00:25:58.660 align:middle line:84% then the quantity of the reduction in labor 00:25:58.660 --> 00:25:59.890 align:middle line:90% will be even larger. 00:25:59.890 --> 00:26:02.425 align:middle line:84% So this quantity reduction here to here is less than here 00:26:02.425 --> 00:26:05.200 align:middle line:84% to here because you don't have the full wage response. 00:26:05.200 --> 00:26:07.900 align:middle line:90% 00:26:07.900 --> 00:26:09.070 align:middle line:90% Wait, is that clear? 00:26:09.070 --> 00:26:13.112 align:middle line:90% Yeah, OK, clear? 00:26:13.112 --> 00:26:13.980 align:middle line:90% OK. 00:26:13.980 --> 00:26:21.670 align:middle line:84% So at some level, I think it's actually 00:26:21.670 --> 00:26:23.170 align:middle line:84% a pretty remarkable set of results 00:26:23.170 --> 00:26:24.628 align:middle line:84% if you actually step back and think 00:26:24.628 --> 00:26:26.080 align:middle line:90% about what's actually going on. 00:26:26.080 --> 00:26:30.250 align:middle line:84% What's going on is that we have these spot-- 00:26:30.250 --> 00:26:32.630 align:middle line:84% you would think of these spot labor markets where-- 00:26:32.630 --> 00:26:35.370 align:middle line:90% 00:26:35.370 --> 00:26:43.060 align:middle line:84% but somehow, these norms about not cutting your nominal wages 00:26:43.060 --> 00:26:47.960 align:middle line:84% seems to be taking place there, too, OK? 00:26:47.960 --> 00:26:50.340 align:middle line:84% So here's-- this is one way of showing this. 00:26:50.340 --> 00:26:51.830 align:middle line:90% They have another paper-- 00:26:51.830 --> 00:26:54.783 align:middle line:84% or actually, [INAUDIBLE] and a different set of coauthors 00:26:54.783 --> 00:26:57.200 align:middle line:84% have another paper-- looking at this experimentally, which 00:26:57.200 --> 00:26:58.430 align:middle line:90% I'll show you in a second. 00:26:58.430 --> 00:27:00.360 align:middle line:84% Yeah, Leslie, you have a question? 00:27:00.360 --> 00:27:00.860 align:middle line:90% No? 00:27:00.860 --> 00:27:04.296 align:middle line:90% Comments, questions? 00:27:04.296 --> 00:27:04.796 align:middle line:90% OK. 00:27:04.796 --> 00:27:12.320 align:middle line:90% 00:27:12.320 --> 00:27:15.260 align:middle line:84% So that says-- it suggested some kind of nominal wage 00:27:15.260 --> 00:27:17.840 align:middle line:90% stickiness. 00:27:17.840 --> 00:27:22.290 align:middle line:84% Another approach could be, suppose 00:27:22.290 --> 00:27:29.190 align:middle line:84% you wanted to test for this lack of market clearing 00:27:29.190 --> 00:27:31.530 align:middle line:90% experimentally. 00:27:31.530 --> 00:27:34.810 align:middle line:90% How would you do that? 00:27:34.810 --> 00:27:37.600 align:middle line:84% Not necessarily this pattern of downward-- 00:27:37.600 --> 00:27:39.830 align:middle line:84% so broadly speaking, the previous paper was saying, 00:27:39.830 --> 00:27:42.940 align:middle line:84% look, we're testing for failures of the market clearing 00:27:42.940 --> 00:27:44.360 align:middle line:84% by testing for one particular one, 00:27:44.360 --> 00:27:45.820 align:middle line:84% which is these patterns of shocks. 00:27:45.820 --> 00:27:47.110 align:middle line:84% More generally, if we want to look 00:27:47.110 --> 00:27:49.360 align:middle line:84% to see whether the wage is not adjusting to the market 00:27:49.360 --> 00:27:51.100 align:middle line:84% clearing level, what else could you-- 00:27:51.100 --> 00:27:52.480 align:middle line:84% suppose you want to do it experimentally. 00:27:52.480 --> 00:27:53.230 align:middle line:90% What could you do? 00:27:53.230 --> 00:27:57.250 align:middle line:90% 00:27:57.250 --> 00:27:57.750 align:middle line:90% John? 00:27:57.750 --> 00:27:58.700 align:middle line:90% 00:27:58.700 --> 00:28:03.052 align:middle line:84% AUDIENCE: You could do a shock to either the labor demand 00:28:03.052 --> 00:28:04.957 align:middle line:90% or the supply, and [INAUDIBLE]. 00:28:04.957 --> 00:28:06.040 align:middle line:90% BEN OLKEN: Exactly, right? 00:28:06.040 --> 00:28:08.770 align:middle line:84% You can just shock either labor demand or labor supply. 00:28:08.770 --> 00:28:11.170 align:middle line:84% So in particular, we just literally-- 00:28:11.170 --> 00:28:13.300 align:middle line:84% if we can somehow experimentally shock this, 00:28:13.300 --> 00:28:16.330 align:middle line:90% we can see what happens. 00:28:16.330 --> 00:28:18.730 align:middle line:84% And in particular, if you think that the wage is 00:28:18.730 --> 00:28:20.980 align:middle line:84% stuck too high, the key prediction 00:28:20.980 --> 00:28:24.130 align:middle line:84% is I can experimentally shock, say, labor demand 00:28:24.130 --> 00:28:26.245 align:middle line:90% and see if the wage moves. 00:28:26.245 --> 00:28:30.000 align:middle line:90% 00:28:30.000 --> 00:28:37.455 align:middle line:84% OK, so that's they're going to do. 00:28:37.455 --> 00:28:39.330 align:middle line:84% So that's literally what they're going to do. 00:28:39.330 --> 00:28:43.390 align:middle line:84% They're going to work in village labor markets, 00:28:43.390 --> 00:28:47.200 align:middle line:84% and they're going to randomly hire 00:28:47.200 --> 00:28:49.300 align:middle line:84% 24% of the labor market in some village to go 00:28:49.300 --> 00:28:51.240 align:middle line:90% do some work in some factory-- 00:28:51.240 --> 00:28:52.890 align:middle line:84% which I think, if I remember correctly, 00:28:52.890 --> 00:28:56.310 align:middle line:84% they were using for some other paper also. 00:28:56.310 --> 00:28:58.050 align:middle line:84% I think they needed a bunch of laborers 00:28:58.050 --> 00:28:59.580 align:middle line:84% to go do some things in a factory 00:28:59.580 --> 00:29:02.220 align:middle line:90% to study some other thing. 00:29:02.220 --> 00:29:04.860 align:middle line:84% I'm not 100% sure of this, but I'm pretty sure it's right. 00:29:04.860 --> 00:29:07.440 align:middle line:84% And so they said, well, if we can just 00:29:07.440 --> 00:29:09.715 align:middle line:84% randomize where we're getting them from, 00:29:09.715 --> 00:29:11.340 align:middle line:84% getting these workers from, we can also 00:29:11.340 --> 00:29:14.205 align:middle line:84% study the effects of this labor [INAUDIBLE] shock. 00:29:14.205 --> 00:29:17.402 align:middle line:84% AUDIENCE: Yeah, how is it not prohibitively expensive to do? 00:29:17.402 --> 00:29:19.610 align:middle line:84% BEN OLKEN: I think they're doing it for like a month, 00:29:19.610 --> 00:29:20.902 align:middle line:90% and they picked small villages. 00:29:20.902 --> 00:29:23.470 align:middle line:90% 00:29:23.470 --> 00:29:24.850 align:middle line:84% Yeah, I mean-- I mean, I actually 00:29:24.850 --> 00:29:30.020 align:middle line:84% don't know the details, but I think it's something like that. 00:29:30.020 --> 00:29:30.660 align:middle line:90% So-- 00:29:30.660 --> 00:29:32.260 align:middle line:84% AUDIENCE: [INAUDIBLE] or something like that? 00:29:32.260 --> 00:29:34.300 align:middle line:84% BEN OLKEN: I don't know where they got the money. 00:29:34.300 --> 00:29:35.450 align:middle line:90% But actually, hold on a second. 00:29:35.450 --> 00:29:36.700 align:middle line:90% So suppose that-- I don't know. 00:29:36.700 --> 00:29:38.740 align:middle line:84% So I don't actually-- you can look in the paper and see what 00:29:38.740 --> 00:29:40.180 align:middle line:90% the numbers are, but if it was-- 00:29:40.180 --> 00:29:41.290 align:middle line:90% I mean, I don't know. 00:29:41.290 --> 00:29:44.860 align:middle line:84% Suppose it was 24% of the labor market. 00:29:44.860 --> 00:29:46.370 align:middle line:90% Suppose the village had-- 00:29:46.370 --> 00:29:49.880 align:middle line:84% it was a small village, a couple hundred people, so 50 people. 00:29:49.880 --> 00:29:52.780 align:middle line:84% 50 people, and they make, I don't 00:29:52.780 --> 00:29:56.440 align:middle line:84% know what the wage is, a couple of bucks a day. 00:29:56.440 --> 00:30:04.060 align:middle line:84% So $100 a day for 50 people times a month 00:30:04.060 --> 00:30:09.370 align:middle line:90% would be $3,000 per village. 00:30:09.370 --> 00:30:11.980 align:middle line:84% And you do that for 50 villages, that's a lot of money, 00:30:11.980 --> 00:30:14.200 align:middle line:90% but it's maybe not like-- 00:30:14.200 --> 00:30:17.640 align:middle line:84% it may not be millions and millions of dollars. 00:30:17.640 --> 00:30:20.250 align:middle line:84% And the other point is, and they also need these people 00:30:20.250 --> 00:30:22.260 align:middle line:90% anyway for some other thing. 00:30:22.260 --> 00:30:24.180 align:middle line:84% So they were getting multiple-- they 00:30:24.180 --> 00:30:27.360 align:middle line:84% needed to hire some labor for some other activity, I think. 00:30:27.360 --> 00:30:30.820 align:middle line:84% And so maybe that's how they did that, I suspect. 00:30:30.820 --> 00:30:33.070 align:middle line:84% Those numbers, by the way, are not the actual numbers. 00:30:33.070 --> 00:30:34.810 align:middle line:84% Those are rough and ready calculations. 00:30:34.810 --> 00:30:36.227 align:middle line:84% I'm sure if you look at the paper, 00:30:36.227 --> 00:30:40.590 align:middle line:84% you could figure out what the rough number is. 00:30:40.590 --> 00:30:41.090 align:middle line:90% OK. 00:30:41.090 --> 00:30:45.690 align:middle line:90% 00:30:45.690 --> 00:30:48.810 align:middle line:84% OK, so that's what they're going to do? 00:30:48.810 --> 00:30:50.680 align:middle line:90% So what are the predictions? 00:30:50.680 --> 00:30:51.920 align:middle line:90% Like what should you look at? 00:30:51.920 --> 00:31:03.328 align:middle line:90% 00:31:03.328 --> 00:31:04.522 align:middle line:90% AUDIENCE: Wage would go up? 00:31:04.522 --> 00:31:05.480 align:middle line:90% BEN OLKEN: What's that? 00:31:05.480 --> 00:31:07.478 align:middle line:90% AUDIENCE: Wage would go up? 00:31:07.478 --> 00:31:09.380 align:middle line:90% BEN OLKEN: OK, so right. 00:31:09.380 --> 00:31:16.760 align:middle line:84% So this is-- so they're going to call this a labor supply shock. 00:31:16.760 --> 00:31:22.370 align:middle line:90% 00:31:22.370 --> 00:31:25.548 align:middle line:84% So it's going to be a negative labor supply shock. 00:31:25.548 --> 00:31:27.590 align:middle line:84% AUDIENCE: Why are they calling it a supply shock? 00:31:27.590 --> 00:31:27.990 align:middle line:90% BEN OLKEN: Huh? 00:31:27.990 --> 00:31:29.680 align:middle line:84% AUDIENCE: Why are they calling it a supply shock? 00:31:29.680 --> 00:31:31.280 align:middle line:84% BEN OLKEN: Because we're pulling laborers-- well, 00:31:31.280 --> 00:31:32.447 align:middle line:90% so you can do it either way. 00:31:32.447 --> 00:31:33.756 align:middle line:90% It actually doesn't matter. 00:31:33.756 --> 00:31:35.550 align:middle line:84% AUDIENCE: So it's a negative supply. 00:31:35.550 --> 00:31:37.733 align:middle line:84% BEN OLKEN: So you can think of there's 00:31:37.733 --> 00:31:39.900 align:middle line:84% a bunch of work in the village, of like farming work 00:31:39.900 --> 00:31:42.630 align:middle line:84% or whatever, And we're going to reduce the labor 00:31:42.630 --> 00:31:45.210 align:middle line:90% supply to that farming work. 00:31:45.210 --> 00:31:46.860 align:middle line:84% I think that's how they think about it. 00:31:46.860 --> 00:31:48.860 align:middle line:84% You could also think about it as the total labor 00:31:48.860 --> 00:31:51.480 align:middle line:84% market in the village and a positive labor demand shock. 00:31:51.480 --> 00:31:52.860 align:middle line:84% I don't think it's really going to matter, 00:31:52.860 --> 00:31:54.652 align:middle line:84% but his is how I think they think about it. 00:31:54.652 --> 00:31:57.810 align:middle line:84% It's like so the labor market for village farm work 00:31:57.810 --> 00:32:00.714 align:middle line:84% has just had a negative labor supply shock. 00:32:00.714 --> 00:32:07.930 align:middle line:84% OK, right, so one view is we went from here to there, 00:32:07.930 --> 00:32:11.730 align:middle line:84% so the wage should go up and the quantity of labor 00:32:11.730 --> 00:32:13.680 align:middle line:90% should go down. 00:32:13.680 --> 00:32:17.190 align:middle line:84% that's if we're in a competitive market. 00:32:17.190 --> 00:32:19.032 align:middle line:90% What else? 00:32:19.032 --> 00:32:20.740 align:middle line:84% And so if it was rationed, what would you 00:32:20.740 --> 00:32:21.657 align:middle line:90% expect to have happen? 00:32:21.657 --> 00:32:30.430 align:middle line:90% 00:32:30.430 --> 00:32:31.450 align:middle line:90% Yeah, Christine? 00:32:31.450 --> 00:32:33.033 align:middle line:84% AUDIENCE: The weight shouldn't change. 00:32:33.033 --> 00:32:35.018 align:middle line:84% BEN OLKEN: Yeah, what about the quantity? 00:32:35.018 --> 00:32:36.560 align:middle line:84% AUDIENCE: It shouldn't change either? 00:32:36.560 --> 00:32:37.280 align:middle line:90% BEN OLKEN: Right. 00:32:37.280 --> 00:32:43.770 align:middle line:84% So if we're rationed up here, we're just 00:32:43.770 --> 00:32:45.150 align:middle line:84% living on the labor demand curve, 00:32:45.150 --> 00:32:50.400 align:middle line:90% and nothing should happen, OK? 00:32:50.400 --> 00:32:54.540 align:middle line:90% So any threats to-- 00:32:54.540 --> 00:32:56.860 align:middle line:84% anything else that you should be worried about here? 00:32:56.860 --> 00:32:58.030 align:middle line:90% Yeah. 00:32:58.030 --> 00:32:59.990 align:middle line:84% AUDIENCE: If your shock is too big, 00:32:59.990 --> 00:33:02.490 align:middle line:84% then you're no longer going to be [INAUDIBLE].. 00:33:02.490 --> 00:33:05.160 align:middle line:84% If you push your labor supply shock too far up-- 00:33:05.160 --> 00:33:08.672 align:middle line:84% BEN OLKEN: Yeah, right, the shock can't be too big. 00:33:08.672 --> 00:33:11.130 align:middle line:84% Because if the shock is too big, then actually, then, it'll 00:33:11.130 --> 00:33:12.190 align:middle line:90% be off the constraint. 00:33:12.190 --> 00:33:14.920 align:middle line:90% So it has to be-- 00:33:14.920 --> 00:33:17.050 align:middle line:84% you're looking for the right size shock 00:33:17.050 --> 00:33:20.680 align:middle line:84% to be detectable, but not so large that it busts-- 00:33:20.680 --> 00:33:22.360 align:middle line:84% it goes over that clearing point. 00:33:22.360 --> 00:33:22.910 align:middle line:90% That's one. 00:33:22.910 --> 00:33:23.410 align:middle line:90% Number two. 00:33:23.410 --> 00:33:25.360 align:middle line:84% Yeah, what else do you think about? 00:33:25.360 --> 00:33:43.040 align:middle line:90% 00:33:43.040 --> 00:33:43.770 align:middle line:90% Anything else? 00:33:43.770 --> 00:33:51.410 align:middle line:90% 00:33:51.410 --> 00:33:56.390 align:middle line:90% So what would happen if-- 00:33:56.390 --> 00:34:02.640 align:middle line:90% 00:34:02.640 --> 00:34:04.976 align:middle line:84% let's see if I can draw this right. 00:34:04.976 --> 00:34:05.710 align:middle line:90% So we have an-- 00:34:05.710 --> 00:34:10.460 align:middle line:90% 00:34:10.460 --> 00:34:14.044 align:middle line:84% and we had, that's our labor supply shock. 00:34:14.044 --> 00:34:17.070 align:middle line:90% 00:34:17.070 --> 00:34:22.484 align:middle line:84% And we had totally inelastic labor demand. 00:34:22.484 --> 00:34:39.449 align:middle line:90% 00:34:39.449 --> 00:34:41.510 align:middle line:90% So what happened there? 00:34:41.510 --> 00:34:48.980 align:middle line:84% This is-- that's right, so what's that? 00:34:48.980 --> 00:34:51.310 align:middle line:84% AUDIENCE: No change in the quantity. 00:34:51.310 --> 00:34:53.110 align:middle line:90% But the wage then goes up? 00:34:53.110 --> 00:34:54.735 align:middle line:84% BEN OLKEN: Yeah, no change in quantity. 00:34:54.735 --> 00:34:55.780 align:middle line:90% The weight goes up. 00:34:55.780 --> 00:34:58.547 align:middle line:84% And likewise, if I drew it totally inelastic, 00:34:58.547 --> 00:35:00.880 align:middle line:84% you can have a change in quantity but no change in wage. 00:35:00.880 --> 00:35:03.742 align:middle line:84% So you might get one but not the other. 00:35:03.742 --> 00:35:05.200 align:middle line:84% So we need to look at both of them. 00:35:05.200 --> 00:35:07.570 align:middle line:84% We need to look at both here, because if we only 00:35:07.570 --> 00:35:10.400 align:middle line:84% looked at one of them, we can't necessarily rule that out. 00:35:10.400 --> 00:35:20.563 align:middle line:84% And what about-- what if the labor supply 00:35:20.563 --> 00:35:21.605 align:middle line:90% curve is totally elastic? 00:35:21.605 --> 00:35:31.680 align:middle line:90% 00:35:31.680 --> 00:35:33.510 align:middle line:90% Leslie, what do you want to say? 00:35:33.510 --> 00:35:34.010 align:middle line:90% No? 00:35:34.010 --> 00:35:35.390 align:middle line:84% You're thinking-- or Patrick, somebody? 00:35:35.390 --> 00:35:36.550 align:middle line:90% AUDIENCE: Nothing happens. 00:35:36.550 --> 00:35:39.980 align:middle line:84% BEN OLKEN: Nothing happens, right? 00:35:39.980 --> 00:35:43.820 align:middle line:90% So I think the one alternative-- 00:35:43.820 --> 00:35:46.370 align:middle line:84% to my mind, one of the alternative things 00:35:46.370 --> 00:35:48.980 align:middle line:84% they have to deal with is, maybe the labor supply 00:35:48.980 --> 00:35:50.558 align:middle line:90% is totally elastic. 00:35:50.558 --> 00:35:52.350 align:middle line:84% Because if labor supply is totally elastic, 00:35:52.350 --> 00:35:56.240 align:middle line:84% then I pull out a bunch of labor, but a bunch of people, 00:35:56.240 --> 00:35:58.580 align:middle line:84% they say, it's a reservation wage, a bunch of people 00:35:58.580 --> 00:36:03.000 align:middle line:84% just enter, and nothing would change. 00:36:03.000 --> 00:36:05.410 align:middle line:90% OK, so just keep that in mind. 00:36:05.410 --> 00:36:08.372 align:middle line:84% And they have-- actually, does anyone 00:36:08.372 --> 00:36:09.830 align:middle line:84% remember, did anyone read the paper 00:36:09.830 --> 00:36:12.915 align:middle line:84% and remember how they talk about labor [INAUDIBLE]?? 00:36:12.915 --> 00:36:15.540 align:middle line:84% They have something in the paper to argue this is not the case, 00:36:15.540 --> 00:36:16.860 align:middle line:84% but I don't actually remember at the moment 00:36:16.860 --> 00:36:18.693 align:middle line:84% exactly why they argue this is not the case. 00:36:18.693 --> 00:36:19.540 align:middle line:90% But it's good. 00:36:19.540 --> 00:36:22.080 align:middle line:84% They do have a paragraph or so on this one. 00:36:22.080 --> 00:36:26.360 align:middle line:84% To my mind, that's the one I would look a little harder at. 00:36:26.360 --> 00:36:29.463 align:middle line:84% OK, the other point is, do you think 00:36:29.463 --> 00:36:31.630 align:middle line:84% the response should be the same throughout the year? 00:36:31.630 --> 00:36:36.870 align:middle line:90% 00:36:36.870 --> 00:36:39.725 align:middle line:90% So here is the key fact. 00:36:39.725 --> 00:36:41.600 align:middle line:84% A key factor about agricultural labor markets 00:36:41.600 --> 00:36:44.597 align:middle line:84% is there are peak periods and non-peak periods. 00:36:44.597 --> 00:36:47.180 align:middle line:84% Peak periods are when we need to plant and we need to harvest, 00:36:47.180 --> 00:36:49.370 align:middle line:84% and people are doing a lot of work then. 00:36:49.370 --> 00:36:52.940 align:middle line:84% And then, the slack period is the rest of the year. 00:36:52.940 --> 00:36:56.380 align:middle line:90% 00:36:56.380 --> 00:36:59.130 align:middle line:84% So do you think there would be likely to be differences 00:36:59.130 --> 00:37:01.968 align:middle line:84% between these two different parts of the year? 00:37:01.968 --> 00:37:03.010 align:middle line:90% Aaron, what do you think? 00:37:03.010 --> 00:37:04.642 align:middle line:84% AUDIENCE: They would be more likely to pick up the effect 00:37:04.642 --> 00:37:06.975 align:middle line:84% that they're they're interested in the peak period, when 00:37:06.975 --> 00:37:10.375 align:middle line:90% there's an active market for-- 00:37:10.375 --> 00:37:11.250 align:middle line:90% BEN OLKEN: Which one? 00:37:11.250 --> 00:37:12.458 align:middle line:90% AUDIENCE: In the peak period. 00:37:12.458 --> 00:37:13.747 align:middle line:90% BEN OLKEN: Why? 00:37:13.747 --> 00:37:15.330 align:middle line:84% AUDIENCE: Because this is when there's 00:37:15.330 --> 00:37:19.030 align:middle line:84% an active market for agricultural labor, 00:37:19.030 --> 00:37:23.810 align:middle line:84% and we are less likely to get an excess labor supply 00:37:23.810 --> 00:37:25.940 align:middle line:90% situation, where taking-- 00:37:25.940 --> 00:37:28.190 align:middle line:84% BEN OLKEN: Sorry, more likely to pick up which effect? 00:37:28.190 --> 00:37:29.650 align:middle line:90% What do you mean? 00:37:29.650 --> 00:37:32.520 align:middle line:84% AUDIENCE: Like if we're in a slack period, 00:37:32.520 --> 00:37:34.020 align:middle line:84% and there's a bunch of excess labor, 00:37:34.020 --> 00:37:37.520 align:middle line:84% and you take 24% of the labor market away, 00:37:37.520 --> 00:37:39.789 align:middle line:84% it's probably not going to lead to a detectable effect 00:37:39.789 --> 00:37:41.832 align:middle line:90% on wages. 00:37:41.832 --> 00:37:42.720 align:middle line:90% BEN OLKEN: OK, sorry. 00:37:42.720 --> 00:37:48.163 align:middle line:84% So going back to our little graphs, 00:37:48.163 --> 00:37:49.330 align:middle line:90% so what do you have in mind? 00:37:49.330 --> 00:37:52.480 align:middle line:90% 00:37:52.480 --> 00:38:00.330 align:middle line:84% So another way to think about it is this is our labor supply, 00:38:00.330 --> 00:38:03.400 align:middle line:84% and this is our labor supply shock. 00:38:03.400 --> 00:38:09.580 align:middle line:84% And maybe this is labor demand in peak, 00:38:09.580 --> 00:38:11.860 align:middle line:84% and maybe I should use different colors. 00:38:11.860 --> 00:38:14.704 align:middle line:84% I'll get those different colors going. 00:38:14.704 --> 00:38:18.496 align:middle line:90% 00:38:18.496 --> 00:38:24.450 align:middle line:84% OK, this is labor demand peak, and I 00:38:24.450 --> 00:38:26.770 align:middle line:90% guess I only have two colors. 00:38:26.770 --> 00:38:29.340 align:middle line:90% Oh, no, we have more colors. 00:38:29.340 --> 00:38:30.300 align:middle line:90% That's excellent. 00:38:30.300 --> 00:38:33.560 align:middle line:90% 00:38:33.560 --> 00:38:34.060 align:middle line:90% Let's sort. 00:38:34.060 --> 00:38:34.560 align:middle line:90% OK. 00:38:34.560 --> 00:38:38.880 align:middle line:90% 00:38:38.880 --> 00:38:41.615 align:middle line:84% And that's labor demand in the slack. 00:38:41.615 --> 00:38:45.500 align:middle line:90% 00:38:45.500 --> 00:38:48.470 align:middle line:90% And this is labor supply. 00:38:48.470 --> 00:38:52.120 align:middle line:90% 00:38:52.120 --> 00:38:54.750 align:middle line:90% And we're going to shock. 00:38:54.750 --> 00:38:57.650 align:middle line:84% And this is our labor supply shock. 00:38:57.650 --> 00:39:01.980 align:middle line:90% OK, so in the peak period-- 00:39:01.980 --> 00:39:05.590 align:middle line:90% 00:39:05.590 --> 00:39:07.630 align:middle line:90% this is all market clearing. 00:39:07.630 --> 00:39:12.700 align:middle line:84% Here, we're going from here to there, and in the slack period, 00:39:12.700 --> 00:39:17.640 align:middle line:84% we're going to go from here to there. 00:39:17.640 --> 00:39:19.310 align:middle line:84% So that's all on the market clearing. 00:39:19.310 --> 00:39:22.400 align:middle line:90% 00:39:22.400 --> 00:39:26.970 align:middle line:90% OK, so now, everyone-- 00:39:26.970 --> 00:39:29.095 align:middle line:84% and by the way, so one assumption 00:39:29.095 --> 00:39:30.470 align:middle line:84% I've made in drawing these graphs 00:39:30.470 --> 00:39:34.063 align:middle line:84% is I haven't tilted the labor supply. 00:39:34.063 --> 00:39:35.480 align:middle line:84% First of all, I've assumed labor-- 00:39:35.480 --> 00:39:39.920 align:middle line:84% sorry, I've assume labor supply is the same in both cases. 00:39:39.920 --> 00:39:44.640 align:middle line:84% I also haven't changed labor demand 00:39:44.640 --> 00:39:47.930 align:middle line:84% to make it more or less elastic in these different periods. 00:39:47.930 --> 00:39:49.680 align:middle line:84% You might also imagine that labor demand-- 00:39:49.680 --> 00:39:51.513 align:middle line:84% I mean, another thing you could also imagine 00:39:51.513 --> 00:39:54.180 align:middle line:84% is maybe labor demand is less elastic in the peak period 00:39:54.180 --> 00:39:55.738 align:middle line:84% because I really, really, really need 00:39:55.738 --> 00:39:57.780 align:middle line:84% to get those crops cleaned now, otherwise they're 00:39:57.780 --> 00:40:00.617 align:middle line:84% going to rot, whereas in the slack period, 00:40:00.617 --> 00:40:02.700 align:middle line:84% I have a couple, some stuff to do around the farm, 00:40:02.700 --> 00:40:04.380 align:middle line:84% but it could get done now, it could get done later-- like, 00:40:04.380 --> 00:40:04.900 align:middle line:90% whatever. 00:40:04.900 --> 00:40:10.830 align:middle line:84% And so that would make it more elastic in the peak period. 00:40:10.830 --> 00:40:12.450 align:middle line:84% But just for the moment, with this, 00:40:12.450 --> 00:40:13.950 align:middle line:84% with this graph like this, does that 00:40:13.950 --> 00:40:17.140 align:middle line:84% give you any thoughts on how things should be changed, 00:40:17.140 --> 00:40:19.390 align:middle line:84% or whether you'd expect anything different in the peak 00:40:19.390 --> 00:40:21.200 align:middle line:90% or the slack periods? 00:40:21.200 --> 00:40:21.880 align:middle line:90% Yeah, Whitney? 00:40:21.880 --> 00:40:22.880 align:middle line:90% Oh, I mean Jenny, sorry. 00:40:22.880 --> 00:40:26.930 align:middle line:84% AUDIENCE: So If you are thinking about rationing 00:40:26.930 --> 00:40:32.680 align:middle line:84% being like a price floor, you're more likely to have that 00:40:32.680 --> 00:40:34.460 align:middle line:90% the case in the slack season. 00:40:34.460 --> 00:40:36.930 align:middle line:90% BEN OLKEN: Yeah, exactly, right? 00:40:36.930 --> 00:40:43.370 align:middle line:84% So if we draw the rationing thing to be, say, 00:40:43.370 --> 00:40:45.170 align:middle line:90% you suppose, for example-- 00:40:45.170 --> 00:40:49.700 align:middle line:90% 00:40:49.700 --> 00:40:51.725 align:middle line:90% sorry, let's draw it-- 00:40:51.725 --> 00:40:53.300 align:middle line:90% ba ba ba bum. 00:40:53.300 --> 00:40:57.695 align:middle line:90% 00:40:57.695 --> 00:40:58.945 align:middle line:90% Let me redraw it a little bit. 00:40:58.945 --> 00:41:05.950 align:middle line:90% 00:41:05.950 --> 00:41:10.710 align:middle line:84% Yeah, let's try it over here, OK? 00:41:10.710 --> 00:41:11.850 align:middle line:90% And let's put the-- 00:41:11.850 --> 00:41:17.514 align:middle line:90% 00:41:17.514 --> 00:41:19.610 align:middle line:90% Wage floor there, that, right? 00:41:19.610 --> 00:41:21.050 align:middle line:90% So if that's the graph-- 00:41:21.050 --> 00:41:22.805 align:middle line:84% hopefully, you all can see my graph-- 00:41:22.805 --> 00:41:26.150 align:middle line:90% 00:41:26.150 --> 00:41:29.300 align:middle line:84% if we have the wage in general is 00:41:29.300 --> 00:41:31.730 align:middle line:84% going to be much higher in the peak season, 00:41:31.730 --> 00:41:34.310 align:middle line:84% and if I just fix a kind of nominal wage floor, 00:41:34.310 --> 00:41:36.590 align:middle line:84% then, in this example, in the peak season, 00:41:36.590 --> 00:41:38.780 align:middle line:84% everything behaves nice and competitively. 00:41:38.780 --> 00:41:40.910 align:middle line:84% The wage goes up, quantity goes down, 00:41:40.910 --> 00:41:42.740 align:middle line:84% because we're above that wage floor. 00:41:42.740 --> 00:41:47.240 align:middle line:84% But in the slack season, we're the whole activity 00:41:47.240 --> 00:41:48.740 align:middle line:84% is happening below that wage floor, 00:41:48.740 --> 00:41:52.100 align:middle line:84% and it looks as if we're rationing. 00:41:52.100 --> 00:41:54.100 align:middle line:84% And this is going to be, basically, what they're 00:41:54.100 --> 00:41:55.450 align:middle line:90% going to argue is happening. 00:41:55.450 --> 00:41:57.990 align:middle line:90% 00:41:57.990 --> 00:42:00.960 align:middle line:90% Is that clear? 00:42:00.960 --> 00:42:05.380 align:middle line:90% OK, so this is how they-- 00:42:05.380 --> 00:42:07.090 align:middle line:84% so this is basically just showing you 00:42:07.090 --> 00:42:08.620 align:middle line:90% this is [INAUDIBLE] true. 00:42:08.620 --> 00:42:09.610 align:middle line:90% So, say that? 00:42:09.610 --> 00:42:10.132 align:middle line:90% OK, fine. 00:42:10.132 --> 00:42:11.590 align:middle line:84% So what they're going to then do is 00:42:11.590 --> 00:42:14.230 align:middle line:84% they're going to then look at the hiring shock-- 00:42:14.230 --> 00:42:16.510 align:middle line:84% that's the negative labor supply reduction-- 00:42:16.510 --> 00:42:20.320 align:middle line:84% interacting with whether you're in the peak period. 00:42:20.320 --> 00:42:22.177 align:middle line:84% And they can measure this just by being 00:42:22.177 --> 00:42:24.010 align:middle line:84% the peak agricultural period in these areas, 00:42:24.010 --> 00:42:25.480 align:middle line:84% or by just the overall employment 00:42:25.480 --> 00:42:29.030 align:middle line:84% level as a measure of how high labor demand is. 00:42:29.030 --> 00:42:34.038 align:middle line:84% So OK, so the first thing they're going to do, 00:42:34.038 --> 00:42:36.580 align:middle line:84% they're going to look at what they call the spillover sample. 00:42:36.580 --> 00:42:38.065 align:middle line:90% So what is this? 00:42:38.065 --> 00:42:39.690 align:middle line:84% So what they do, and this is actually-- 00:42:39.690 --> 00:42:42.840 align:middle line:90% I think this is kind of clever-- 00:42:42.840 --> 00:42:48.340 align:middle line:84% so what they do is they have to keep track of who's who 00:42:48.340 --> 00:42:51.050 align:middle line:84% in these different labor markets. 00:42:51.050 --> 00:42:52.810 align:middle line:84% So what they do is, before they start, 00:42:52.810 --> 00:42:54.400 align:middle line:84% they recruit a list of workers who 00:42:54.400 --> 00:42:57.150 align:middle line:84% are interested in their factory jobs. 00:42:57.150 --> 00:42:59.760 align:middle line:84% That's going to be the potential treatment set. 00:42:59.760 --> 00:43:02.935 align:middle line:84% And they randomly pick which villagers 00:43:02.935 --> 00:43:04.060 align:middle line:90% they're going to hire from. 00:43:04.060 --> 00:43:05.477 align:middle line:84% And within villages, they randomly 00:43:05.477 --> 00:43:07.890 align:middle line:90% pick people from this list. 00:43:07.890 --> 00:43:11.440 align:middle line:84% The spillover sample is everyone who 00:43:11.440 --> 00:43:15.310 align:middle line:84% is not on this list, who did not say they were interested in one 00:43:15.310 --> 00:43:17.560 align:middle line:90% of these jobs. 00:43:17.560 --> 00:43:19.690 align:middle line:84% And the point that I want to make 00:43:19.690 --> 00:43:22.420 align:middle line:84% is, because they recruited this list ex ante, 00:43:22.420 --> 00:43:25.120 align:middle line:84% before they randomized, they know 00:43:25.120 --> 00:43:28.150 align:middle line:84% who was in the spillover sample in both treatment and control 00:43:28.150 --> 00:43:29.800 align:middle line:90% villages kind of identically. 00:43:29.800 --> 00:43:32.890 align:middle line:84% Because step one is I split the villages into, say, 00:43:32.890 --> 00:43:35.140 align:middle line:84% this half of the villages, if you're a treatment area, 00:43:35.140 --> 00:43:37.180 align:middle line:84% we're going to give you factory work. 00:43:37.180 --> 00:43:39.760 align:middle line:84% This half of the village, if you were in the treatment area, 00:43:39.760 --> 00:43:41.650 align:middle line:84% we're not going to give you factory work. 00:43:41.650 --> 00:43:44.860 align:middle line:84% And we know who you are because you identify yourself 00:43:44.860 --> 00:43:46.000 align:middle line:90% beforehand. 00:43:46.000 --> 00:43:49.370 align:middle line:84% And then, when I randomize you into the experiment later on, 00:43:49.370 --> 00:43:52.300 align:middle line:84% it's then totally OK for me to just look at this set of people 00:43:52.300 --> 00:43:57.400 align:middle line:84% and in control treatment areas as the rest of the labor market 00:43:57.400 --> 00:43:58.960 align:middle line:84% because I identified them identically 00:43:58.960 --> 00:44:00.190 align:middle line:90% in treatment control areas. 00:44:00.190 --> 00:44:03.700 align:middle line:84% If I hadn't done this step, it would still 00:44:03.700 --> 00:44:05.450 align:middle line:84% be OK to look at the total labor market, 00:44:05.450 --> 00:44:06.992 align:middle line:84% but I couldn't just specifically look 00:44:06.992 --> 00:44:08.992 align:middle line:84% at what's happening to this other set of workers 00:44:08.992 --> 00:44:10.990 align:middle line:84% because I wouldn't know, in the treatment area, 00:44:10.990 --> 00:44:12.820 align:middle line:84% in the control areas, who would have been the people who would 00:44:12.820 --> 00:44:13.840 align:middle line:84% have taken over the factory job and who 00:44:13.840 --> 00:44:15.100 align:middle line:90% wouldn't have been the people. 00:44:15.100 --> 00:44:16.610 align:middle line:90% You see what I'm saying? 00:44:16.610 --> 00:44:20.860 align:middle line:84% So the fact that this is all very precise and done ex ante 00:44:20.860 --> 00:44:23.565 align:middle line:90% is very helpful. 00:44:23.565 --> 00:44:24.065 align:middle line:90% Clear? 00:44:24.065 --> 00:44:28.283 align:middle line:90% 00:44:28.283 --> 00:44:30.450 align:middle line:84% So the first thing to look at, the spillover sample. 00:44:30.450 --> 00:44:33.747 align:middle line:90% 00:44:33.747 --> 00:44:35.580 align:middle line:84% One thing you should know about, by the way, 00:44:35.580 --> 00:44:39.470 align:middle line:84% is that rural wages often take the form of a bit of wage 00:44:39.470 --> 00:44:40.963 align:middle line:90% and a bit of in-kind. 00:44:40.963 --> 00:44:43.130 align:middle line:84% They'll give you some cash, and then, maybe, they'll 00:44:43.130 --> 00:44:45.118 align:middle line:84% give you lunch, or something like that. 00:44:45.118 --> 00:44:46.160 align:middle line:90% So you actually have to-- 00:44:46.160 --> 00:44:50.150 align:middle line:84% and actually, for a rural laborer 00:44:50.150 --> 00:44:53.840 align:middle line:84% making subsistence wage, lunch can be a non-trivial fraction 00:44:53.840 --> 00:44:54.950 align:middle line:90% of the total compensation. 00:44:54.950 --> 00:44:57.605 align:middle line:84% So you have to be careful to measure that, too. 00:44:57.605 --> 00:44:59.330 align:middle line:90% Like lunch and snacks. 00:44:59.330 --> 00:45:00.950 align:middle line:84% I feel like in one case, it was like-- 00:45:00.950 --> 00:45:03.555 align:middle line:84% I said it was like lunch, a snack, 00:45:03.555 --> 00:45:04.680 align:middle line:90% and a couple of cigarettes. 00:45:04.680 --> 00:45:07.850 align:middle line:84% It was like that stuff actually can be non-trivial relative 00:45:07.850 --> 00:45:10.398 align:middle line:84% to the whole conversation, so you have to include it too. 00:45:10.398 --> 00:45:12.440 align:middle line:84% OK, so in the spillover sample, what do you find? 00:45:12.440 --> 00:45:15.500 align:middle line:84% So some want to interpret what we're finding here? 00:45:15.500 --> 00:45:28.250 align:middle line:90% 00:45:28.250 --> 00:45:30.997 align:middle line:84% AUDIENCE: The shock has no effect in the-- a period 00:45:30.997 --> 00:45:33.398 align:middle line:90% of very little effect and-- 00:45:33.398 --> 00:45:34.690 align:middle line:90% BEN OLKEN: What are you saying? 00:45:34.690 --> 00:45:35.815 align:middle line:90% What, sorry, what was that? 00:45:35.815 --> 00:45:37.420 align:middle line:84% AUDIENCE: The shock, the supply shock, 00:45:37.420 --> 00:45:41.322 align:middle line:84% has very little impact on wages [INAUDIBLE].. 00:45:41.322 --> 00:45:44.280 align:middle line:90% 00:45:44.280 --> 00:45:45.400 align:middle line:90% BEN OLKEN: Right. 00:45:45.400 --> 00:45:47.317 align:middle line:84% I don't remember why it's called [INAUDIBLE].. 00:45:47.317 --> 00:45:50.162 align:middle line:90% Let's just call it [INAUDIBLE]. 00:45:50.162 --> 00:45:53.120 align:middle line:90% So what do you think of that? 00:45:53.120 --> 00:45:55.087 align:middle line:84% AUDIENCE: It seems consistent with-- 00:45:55.087 --> 00:45:56.920 align:middle line:84% BEN OLKEN: Our nice little graph over there? 00:45:56.920 --> 00:45:57.370 align:middle line:90% AUDIENCE: Yeah. 00:45:57.370 --> 00:45:58.703 align:middle line:90% BEN OLKEN: Yeah, exactly, right? 00:45:58.703 --> 00:46:01.190 align:middle line:90% OK, so that's on wages. 00:46:01.190 --> 00:46:04.520 align:middle line:84% And so exactly, that's exactly what You would-- 00:46:04.520 --> 00:46:06.510 align:middle line:84% that's exactly this point over here, 00:46:06.510 --> 00:46:08.390 align:middle line:84% which is in the non-peak period, it 00:46:08.390 --> 00:46:11.360 align:middle line:84% looks like they're constrained, like the wage is unaffected. 00:46:11.360 --> 00:46:14.630 align:middle line:84% In the peak period, the wage is affected. 00:46:14.630 --> 00:46:18.020 align:middle line:84% And the second thing is we want to look at employment. 00:46:18.020 --> 00:46:21.380 align:middle line:84% So this is still the spillover sample. 00:46:21.380 --> 00:46:22.728 align:middle line:90% Ignore this bullet point. 00:46:22.728 --> 00:46:24.770 align:middle line:84% I changed the graph and didn't change the bullet, 00:46:24.770 --> 00:46:25.700 align:middle line:90% so just ignore that. 00:46:25.700 --> 00:46:28.420 align:middle line:90% 00:46:28.420 --> 00:46:29.280 align:middle line:90% So this is the-- 00:46:29.280 --> 00:46:33.900 align:middle line:90% 00:46:33.900 --> 00:46:36.710 align:middle line:84% OK, sorry, so what do you make of this? 00:46:36.710 --> 00:46:38.835 align:middle line:84% This is employment from the spillover sample. 00:46:38.835 --> 00:46:51.737 align:middle line:90% 00:46:51.737 --> 00:46:54.070 align:middle line:84% AUDIENCE: Decrease in employment during the peak season, 00:46:54.070 --> 00:46:56.710 align:middle line:84% and less of a decrease in employment 00:46:56.710 --> 00:46:58.848 align:middle line:90% during the offseason. 00:46:58.848 --> 00:47:00.140 align:middle line:90% BEN OLKEN: Say it again, sorry. 00:47:00.140 --> 00:47:00.430 align:middle line:90% I wasn't listening. 00:47:00.430 --> 00:47:01.055 align:middle line:90% Say that again? 00:47:01.055 --> 00:47:11.120 align:middle line:84% AUDIENCE: So the hiring shock in the spillover oversample causes 00:47:11.120 --> 00:47:15.830 align:middle line:84% less of a decrease in employment during the off-peak season 00:47:15.830 --> 00:47:18.530 align:middle line:84% than during the semi-peak season. 00:47:18.530 --> 00:47:21.650 align:middle line:90% BEN OLKEN: Not quite. 00:47:21.650 --> 00:47:24.230 align:middle line:84% So I think-- so you're right that there's an interaction 00:47:24.230 --> 00:47:26.795 align:middle line:84% effect, but the main effect you shouldn't-- 00:47:26.795 --> 00:47:30.060 align:middle line:84% I think you're interpreting the wrong way. 00:47:30.060 --> 00:47:34.770 align:middle line:84% So it's positive in the off-peak season. 00:47:34.770 --> 00:47:36.290 align:middle line:90% AUDIENCE: Yes. 00:47:36.290 --> 00:47:37.490 align:middle line:90% BEN OLKEN: Right? 00:47:37.490 --> 00:47:39.980 align:middle line:84% So employment goes up in the off-peak season, 00:47:39.980 --> 00:47:43.835 align:middle line:84% and in the peak season, it doesn't change. 00:47:43.835 --> 00:47:44.460 align:middle line:90% AUDIENCE: Yeah. 00:47:44.460 --> 00:47:45.877 align:middle line:84% BEN OLKEN: Does everyone see that? 00:47:45.877 --> 00:47:49.080 align:middle line:84% So for the off-peak season, it's just positive, whatever, 00:47:49.080 --> 00:47:49.980 align:middle line:90% 6%, 10%. 00:47:49.980 --> 00:47:53.340 align:middle line:84% For the non-peak season, it's basically 0 00:47:53.340 --> 00:47:55.060 align:middle line:90% because we add these two things. 00:47:55.060 --> 00:47:56.560 align:middle line:84% I'm sorry, for the peak season, it's 00:47:56.560 --> 00:48:00.166 align:middle line:84% basically 0 because we add these two things, right? 00:48:00.166 --> 00:48:05.300 align:middle line:84% OK, so what is that telling us vis a vis the model over there? 00:48:05.300 --> 00:48:07.285 align:middle line:84% Is that consistent, not consistent? 00:48:07.285 --> 00:48:22.340 align:middle line:90% 00:48:22.340 --> 00:48:24.290 align:middle line:90% Let me show you one more thing. 00:48:24.290 --> 00:48:26.570 align:middle line:84% One more piece, and then we'll put it all together. 00:48:26.570 --> 00:48:28.070 align:middle line:84% This one is actually a little harder 00:48:28.070 --> 00:48:29.420 align:middle line:84% to think about because this is the spillover. 00:48:29.420 --> 00:48:30.800 align:middle line:84% This is what's happening just to the spillover-- 00:48:30.800 --> 00:48:32.540 align:middle line:84% I'm sorry, I almost lasered you guys-- 00:48:32.540 --> 00:48:36.538 align:middle line:84% to the spillover sample, the people over there who were not, 00:48:36.538 --> 00:48:38.330 align:middle line:84% potentially, included in the hiring sample. 00:48:38.330 --> 00:48:43.400 align:middle line:90% 00:48:43.400 --> 00:48:44.800 align:middle line:90% This is total employment. 00:48:44.800 --> 00:48:47.300 align:middle line:84% So now, we're just going to pool the whole village together. 00:48:47.300 --> 00:48:51.010 align:middle line:90% 00:48:51.010 --> 00:48:52.305 align:middle line:90% And what do we learn from this? 00:48:52.305 --> 00:49:06.300 align:middle line:90% 00:49:06.300 --> 00:49:07.020 align:middle line:90% Patrick? 00:49:07.020 --> 00:49:11.500 align:middle line:84% AUDIENCE: That the individuals from this spillover sample-- 00:49:11.500 --> 00:49:12.750 align:middle line:90% BEN OLKEN: This is everything. 00:49:12.750 --> 00:49:13.440 align:middle line:90% This is everybody. 00:49:13.440 --> 00:49:14.982 align:middle line:84% AUDIENCE: Right, right, but does this 00:49:14.982 --> 00:49:17.540 align:middle line:84% mean that individuals in this spillover sample 00:49:17.540 --> 00:49:21.330 align:middle line:84% increase their level of supply [INAUDIBLE]?? 00:49:21.330 --> 00:49:22.260 align:middle line:90% BEN OLKEN: Yeah, OK. 00:49:22.260 --> 00:49:24.640 align:middle line:84% First, before we try to interpret what this means, 00:49:24.640 --> 00:49:25.890 align:middle line:90% what is this table telling us? 00:49:25.890 --> 00:49:29.683 align:middle line:90% 00:49:29.683 --> 00:49:31.100 align:middle line:84% It may have been easier, actually, 00:49:31.100 --> 00:49:33.480 align:middle line:84% if I started with this one before the spillover sample. 00:49:33.480 --> 00:49:35.300 align:middle line:84% So let's take the weighted results-- 00:49:35.300 --> 00:49:36.050 align:middle line:90% Fine. 00:49:36.050 --> 00:49:39.410 align:middle line:84% No effect in the slack period, big positive effect 00:49:39.410 --> 00:49:40.333 align:middle line:90% in the peak period. 00:49:40.333 --> 00:49:42.000 align:middle line:84% And now, let's look at total quantities, 00:49:42.000 --> 00:49:42.833 align:middle line:90% and what do we find? 00:49:42.833 --> 00:49:48.683 align:middle line:90% 00:49:48.683 --> 00:49:49.725 align:middle line:90% Hazel, what do you think? 00:49:49.725 --> 00:49:53.460 align:middle line:84% AUDIENCE: No effect in the off-peak, and then [INAUDIBLE] 00:49:53.460 --> 00:49:54.030 align:middle line:90% goes down. 00:49:54.030 --> 00:49:55.155 align:middle line:90% BEN OLKEN: And what, sorry? 00:49:55.155 --> 00:49:57.550 align:middle line:84% AUDIENCE: Total quantity goes down in peak season. 00:49:57.550 --> 00:49:59.290 align:middle line:84% BEN OLKEN: Right, total right quantity 00:49:59.290 --> 00:50:01.390 align:middle line:84% goes down in the peak season, right? 00:50:01.390 --> 00:50:05.300 align:middle line:90% OK, so what is this telling us? 00:50:05.300 --> 00:50:06.942 align:middle line:84% Is that consistent with that or not? 00:50:06.942 --> 00:50:08.900 align:middle line:84% AUDIENCE: That's consistent with the left-hand. 00:50:08.900 --> 00:50:09.980 align:middle line:84% BEN OLKEN: Exactly, it's exactly consistent 00:50:09.980 --> 00:50:11.040 align:middle line:90% with left-handed graph. 00:50:11.040 --> 00:50:13.360 align:middle line:84% So then, so right, so what do we see? 00:50:13.360 --> 00:50:16.700 align:middle line:90% And so in the non-peak period-- 00:50:16.700 --> 00:50:17.930 align:middle line:90% we're stuck over here. 00:50:17.930 --> 00:50:19.400 align:middle line:90% Nothing happens. 00:50:19.400 --> 00:50:23.568 align:middle line:84% In the peak period, it's going up. 00:50:23.568 --> 00:50:25.860 align:middle line:84% So it seems that these two results are the key results. 00:50:25.860 --> 00:50:26.450 align:middle line:90% So why? 00:50:26.450 --> 00:50:28.520 align:middle line:84% What is the spillover result telling us? 00:50:28.520 --> 00:50:30.948 align:middle line:84% Well, you might have just thought, 00:50:30.948 --> 00:50:32.990 align:middle line:84% something weird is happening in the labor market. 00:50:32.990 --> 00:50:35.510 align:middle line:90% Nothing is going on. 00:50:35.510 --> 00:50:37.520 align:middle line:84% Nothing doing in the slack period. 00:50:37.520 --> 00:50:40.070 align:middle line:84% What's nice about the spillover result is that actually, 00:50:40.070 --> 00:50:43.653 align:middle line:84% no, no, no, in the spillover sample, in the non-peak period, 00:50:43.653 --> 00:50:44.820 align:middle line:90% what do we expect to happen? 00:50:44.820 --> 00:50:46.320 align:middle line:84% Well we expect, we just took a bunch 00:50:46.320 --> 00:50:48.960 align:middle line:84% of people out of the labor market-- 00:50:48.960 --> 00:50:51.130 align:middle line:90% out of this rural labor market. 00:50:51.130 --> 00:50:53.850 align:middle line:90% Some of them were working. 00:50:53.850 --> 00:50:57.690 align:middle line:84% The total quantity, the total people working in this labor 00:50:57.690 --> 00:51:04.070 align:middle line:90% market, should be unchanged. 00:51:04.070 --> 00:51:05.990 align:middle line:84% Therefore, we have to pull in some 00:51:05.990 --> 00:51:08.700 align:middle line:84% of the spillover people who have to come and take those jobs. 00:51:08.700 --> 00:51:10.580 align:middle line:84% So that's the final piece that's showing you 00:51:10.580 --> 00:51:14.410 align:middle line:84% that something's actually happening in this period. 00:51:14.410 --> 00:51:16.250 align:middle line:90% Is that clear? 00:51:16.250 --> 00:51:17.120 align:middle line:90% Yeah, Whitney. 00:51:17.120 --> 00:51:18.620 align:middle line:84% AUDIENCE: So when you're actually 00:51:18.620 --> 00:51:20.450 align:middle line:84% designing this experiment, and you 00:51:20.450 --> 00:51:22.760 align:middle line:84% want to get this effect again, how 00:51:22.760 --> 00:51:26.360 align:middle line:84% do you know the 24% is the number of people you 00:51:26.360 --> 00:51:28.490 align:middle line:90% need to be pulling to get this? 00:51:28.490 --> 00:51:30.690 align:middle line:90% BEN OLKEN: Good question! 00:51:30.690 --> 00:51:32.990 align:middle line:84% So how do you know that this is a big enough effect? 00:51:32.990 --> 00:51:35.700 align:middle line:90% 00:51:35.700 --> 00:51:38.600 align:middle line:84% Well, so the answer is you would do a power calculation. 00:51:38.600 --> 00:51:42.230 align:middle line:84% So you would say like- so you don't know. 00:51:42.230 --> 00:51:44.390 align:middle line:90% I mean, you don't fully know. 00:51:44.390 --> 00:51:46.760 align:middle line:84% And actually, I would love to hear the author's answer 00:51:46.760 --> 00:51:49.100 align:middle line:84% to that question of what exactly they did. 00:51:49.100 --> 00:51:51.950 align:middle line:84% I suspect the truth was, it the big enough shock 00:51:51.950 --> 00:51:53.030 align:middle line:90% they could afford. 00:51:53.030 --> 00:51:57.740 align:middle line:84% [LAUGHS] If you actually asked, I suspect 00:51:57.740 --> 00:51:58.880 align:middle line:90% that's what's going on. 00:51:58.880 --> 00:52:00.797 align:middle line:84% But I think you can actually do a little more. 00:52:00.797 --> 00:52:04.500 align:middle line:84% So you can say, well, look, you can calculate-- 00:52:04.500 --> 00:52:09.600 align:middle line:90% 00:52:09.600 --> 00:52:11.730 align:middle line:90% so what can you do? 00:52:11.730 --> 00:52:19.425 align:middle line:84% So you can calculate, if you have 00:52:19.425 --> 00:52:21.800 align:middle line:84% some estimate for the slope of the labor supply and labor 00:52:21.800 --> 00:52:29.000 align:middle line:84% demand curves, you can say, if I shock labor supply by 24%, how 00:52:29.000 --> 00:52:31.280 align:middle line:84% large an effect would I expect on wages, 00:52:31.280 --> 00:52:33.950 align:middle line:84% and how large an effect I would expect on quantities? 00:52:33.950 --> 00:52:41.345 align:middle line:84% That's just, you have these elasticities, 00:52:41.345 --> 00:52:43.150 align:middle line:84% so you know the percent change in-- 00:52:43.150 --> 00:52:46.100 align:middle line:90% 00:52:46.100 --> 00:52:46.650 align:middle line:90% so whatever. 00:52:46.650 --> 00:52:48.860 align:middle line:84% If I have a 24% reduction in labor supply 00:52:48.860 --> 00:52:55.010 align:middle line:84% and a labor demand elasticity of whatever, 00:52:55.010 --> 00:52:57.680 align:middle line:84% I can multiply the labor supply and elasticity 00:52:57.680 --> 00:53:01.310 align:middle line:84% times the [INAUDIBLE],, divide it, 00:53:01.310 --> 00:53:03.890 align:middle line:84% and figure out how big an expected 00:53:03.890 --> 00:53:05.482 align:middle line:90% change you're going to get. 00:53:05.482 --> 00:53:07.690 align:middle line:84% So what I would do if I was designing this experiment 00:53:07.690 --> 00:53:10.660 align:middle line:84% is I would, A, come up with my favorite labor 00:53:10.660 --> 00:53:13.660 align:middle line:84% supply-demand elasticity from some other estimate, 00:53:13.660 --> 00:53:16.565 align:middle line:84% or at least try to get it from somewhere. 00:53:16.565 --> 00:53:18.190 align:middle line:84% I guess you also need the labor supply. 00:53:18.190 --> 00:53:19.565 align:middle line:84% You need both of these estimates. 00:53:19.565 --> 00:53:22.990 align:middle line:84% So take your favorite of these estimates. 00:53:22.990 --> 00:53:25.210 align:middle line:90% You would-- oh, no, sorry. 00:53:25.210 --> 00:53:28.510 align:middle line:84% Mostly, you just need the labor demand elasticity. 00:53:28.510 --> 00:53:30.610 align:middle line:84% I would take my favorite estimate. 00:53:30.610 --> 00:53:34.210 align:middle line:84% I would multiply it by 24% to figure out 00:53:34.210 --> 00:53:36.220 align:middle line:84% how big a change in wages or quantities 00:53:36.220 --> 00:53:38.133 align:middle line:84% I would likely to be expected to get. 00:53:38.133 --> 00:53:39.550 align:middle line:84% And the second thing I would do is 00:53:39.550 --> 00:53:41.110 align:middle line:84% I would then do a power calculation 00:53:41.110 --> 00:53:45.160 align:middle line:84% and say, well, I'm going to measure wages 00:53:45.160 --> 00:53:48.380 align:middle line:84% in this village in quantities in the following way. 00:53:48.380 --> 00:53:50.680 align:middle line:84% I can get some estimate from some other data sets 00:53:50.680 --> 00:53:52.810 align:middle line:84% that I've done of what the standard errors are 00:53:52.810 --> 00:53:56.010 align:middle line:90% likely to be in this regression. 00:53:56.010 --> 00:54:00.250 align:middle line:84% And I can put those two things together and say, 00:54:00.250 --> 00:54:04.650 align:middle line:84% I expect to find a difference in the wages between these two 00:54:04.650 --> 00:54:07.890 align:middle line:84% periods of, say, 7% based on my best 00:54:07.890 --> 00:54:11.118 align:middle line:84% estimates of those elasticities and the size of the shock. 00:54:11.118 --> 00:54:12.660 align:middle line:84% What's the probability I could detect 00:54:12.660 --> 00:54:15.077 align:middle line:84% that given the expected standard errors I'm going to find? 00:54:15.077 --> 00:54:17.520 align:middle line:84% And that's what a power calculation exactly does. 00:54:17.520 --> 00:54:20.040 align:middle line:90% So that's how you would do it. 00:54:20.040 --> 00:54:22.920 align:middle line:90% Now, that requires two things. 00:54:22.920 --> 00:54:24.860 align:middle line:84% It requires having the correct estimate 00:54:24.860 --> 00:54:27.260 align:middle line:84% of those elasticities, or at least 00:54:27.260 --> 00:54:28.640 align:middle line:84% something that's in the ballpark. 00:54:28.640 --> 00:54:31.015 align:middle line:84% But even if you don't get it exactly right, you can get-- 00:54:31.015 --> 00:54:33.260 align:middle line:84% the correlations are never perfect, 00:54:33.260 --> 00:54:36.080 align:middle line:84% but you can at least get it in the rough ballpark. 00:54:36.080 --> 00:54:37.850 align:middle line:84% And it also requires having some estimate 00:54:37.850 --> 00:54:39.510 align:middle line:84% of what the standard errors are going to look like. 00:54:39.510 --> 00:54:41.927 align:middle line:84% And that's going to depend on the data generating process, 00:54:41.927 --> 00:54:44.420 align:middle line:84% and what your sample looks like, and so on, and so forth. 00:54:44.420 --> 00:54:47.640 align:middle line:84% If you've done other samples of wages in these areas, 00:54:47.640 --> 00:54:50.060 align:middle line:84% you can do a pilot survey to get some of that data, 00:54:50.060 --> 00:54:51.395 align:middle line:90% and you use that. 00:54:51.395 --> 00:54:54.500 align:middle line:84% So that's actually how you would do that. 00:54:54.500 --> 00:54:56.285 align:middle line:84% Sorry, Paolo, got another question? 00:54:56.285 --> 00:54:58.410 align:middle line:84% AUDIENCE: I just have a very quick, basic question. 00:54:58.410 --> 00:55:03.015 align:middle line:84% What is the employment level here in these tables? 00:55:03.015 --> 00:55:04.640 align:middle line:84% BEN OLKEN: So actually I will tell you. 00:55:04.640 --> 00:55:05.960 align:middle line:84% I don't remember exactly, but I believe 00:55:05.960 --> 00:55:07.793 align:middle line:84% it's a measure of total aggregate employment 00:55:07.793 --> 00:55:12.073 align:middle line:84% in the village in that month as a way of-- 00:55:12.073 --> 00:55:14.240 align:middle line:84% as an alternative measure of measuring whether we're 00:55:14.240 --> 00:55:15.650 align:middle line:90% in peak season or not. 00:55:15.650 --> 00:55:18.200 align:middle line:84% Basically, so one is we can define peak season 00:55:18.200 --> 00:55:19.160 align:middle line:90% based on season. 00:55:19.160 --> 00:55:22.327 align:middle line:84% The other is we can define it based on q, [INAUDIBLE] of q. 00:55:22.327 --> 00:55:24.410 align:middle line:84% I don't remember the exact definition I used of q, 00:55:24.410 --> 00:55:27.270 align:middle line:90% but that's the idea. 00:55:27.270 --> 00:55:27.770 align:middle line:90% Jenny? 00:55:27.770 --> 00:55:29.687 align:middle line:84% AUDIENCE: I think I'm still confused about why 00:55:29.687 --> 00:55:32.130 align:middle line:84% it's important to distinguish the [INAUDIBLE] group from 00:55:32.130 --> 00:55:34.530 align:middle line:84% people who don't get the job but were interested, 00:55:34.530 --> 00:55:36.150 align:middle line:90% unless you're worried about-- 00:55:36.150 --> 00:55:37.442 align:middle line:90% BEN OLKEN: Composition effects. 00:55:37.442 --> 00:55:45.750 align:middle line:90% 00:55:45.750 --> 00:55:48.210 align:middle line:84% So for wages-- wages is the easiest case. 00:55:48.210 --> 00:55:50.190 align:middle line:90% Wages, it's easier. 00:55:50.190 --> 00:55:51.540 align:middle line:90% You see why you want to do that? 00:55:51.540 --> 00:55:53.430 align:middle line:84% Imagine that we have good workers and bad workers, 00:55:53.430 --> 00:55:55.770 align:middle line:84% good agricultural workers and bad agricultural workers. 00:55:55.770 --> 00:55:58.050 align:middle line:84% So good agricultural workers are the ones who 00:55:58.050 --> 00:56:01.985 align:middle line:90% are doing the job in general. 00:56:01.985 --> 00:56:04.360 align:middle line:84% Imagine I just take a bunch of those agricultural workers 00:56:04.360 --> 00:56:06.432 align:middle line:90% and pluck them out. 00:56:06.432 --> 00:56:08.390 align:middle line:84% They're going to be replaced with lower-quality 00:56:08.390 --> 00:56:09.890 align:middle line:90% agricultural workers. 00:56:09.890 --> 00:56:12.950 align:middle line:84% If I'm measuring the wage, not particularly-- 00:56:12.950 --> 00:56:14.390 align:middle line:84% like the wage I'm going to measure 00:56:14.390 --> 00:56:16.520 align:middle line:84% might be your average marginal product. 00:56:16.520 --> 00:56:18.320 align:middle line:84% If I replace good workers with bad workers, 00:56:18.320 --> 00:56:20.360 align:middle line:84% the wage is going to fall, not because of these effects, 00:56:20.360 --> 00:56:22.610 align:middle line:84% but just because I've moved to having worse quality 00:56:22.610 --> 00:56:27.290 align:middle line:84% workers who are being paid based on their lower 00:56:27.290 --> 00:56:28.460 align:middle line:90% marginal productivity. 00:56:28.460 --> 00:56:30.620 align:middle line:84% So the reason for doing it on the spillover sample 00:56:30.620 --> 00:56:32.750 align:middle line:84% is we're conditioning on the same set of people. 00:56:32.750 --> 00:56:36.557 align:middle line:90% 00:56:36.557 --> 00:56:38.140 align:middle line:84% It's not perfect, actually, now that I 00:56:38.140 --> 00:56:41.350 align:middle line:84% think about it because you're getting different-- 00:56:41.350 --> 00:56:42.860 align:middle line:84% it's among people who are working, 00:56:42.860 --> 00:56:43.990 align:middle line:84% so you are still getting some selection. 00:56:43.990 --> 00:56:45.448 align:middle line:84% But I think it's a little bit less. 00:56:45.448 --> 00:56:49.250 align:middle line:84% I think that's because it's still conditional. 00:56:49.250 --> 00:56:52.030 align:middle line:84% So if everyone was working, and it 00:56:52.030 --> 00:56:53.830 align:middle line:84% was the same people working both periods, 00:56:53.830 --> 00:56:55.060 align:middle line:90% it would be totally clean. 00:56:55.060 --> 00:56:58.627 align:middle line:84% Now that I think about the explanation I just gave you, 00:56:58.627 --> 00:57:00.960 align:middle line:84% you still could have some selection within the spillover 00:57:00.960 --> 00:57:03.060 align:middle line:84% sample, but I think it's somewhat less. 00:57:03.060 --> 00:57:04.560 align:middle line:84% AUDIENCE: So the idea is that people 00:57:04.560 --> 00:57:07.350 align:middle line:84% who are interested in applying for this external job 00:57:07.350 --> 00:57:09.660 align:middle line:84% are more likely to already be in the labor force 00:57:09.660 --> 00:57:10.980 align:middle line:90% and have selected [INAUDIBLE]? 00:57:10.980 --> 00:57:13.273 align:middle line:84% BEN OLKEN: No, it's that there's a mechanical selection 00:57:13.273 --> 00:57:15.190 align:middle line:84% effect if we pull them out of the labor force. 00:57:15.190 --> 00:57:19.210 align:middle line:84% The selection effect is much larger for that sample. 00:57:19.210 --> 00:57:21.620 align:middle line:84% Because I'm literally taking these people out, 00:57:21.620 --> 00:57:24.895 align:middle line:90% and so there's like a large-- 00:57:24.895 --> 00:57:27.460 align:middle line:84% I think the mechanical effect of who's in the labor force 00:57:27.460 --> 00:57:30.090 align:middle line:90% is much larger over there. 00:57:30.090 --> 00:57:31.298 align:middle line:90% But I agree, actually. 00:57:31.298 --> 00:57:33.090 align:middle line:84% Now that I say it, it's not perfect either. 00:57:33.090 --> 00:57:34.757 align:middle line:84% But I think that's, at least, the theory 00:57:34.757 --> 00:57:35.910 align:middle line:90% for how we think about it. 00:57:35.910 --> 00:57:36.570 align:middle line:90% Patrick? 00:57:36.570 --> 00:57:38.884 align:middle line:84% AUDIENCE: Why not introduce a variation of degree 00:57:38.884 --> 00:57:42.120 align:middle line:90% of [INAUDIBLE]? 00:57:42.120 --> 00:57:44.730 align:middle line:84% BEN OLKEN: Oh, OK, so tell me why you would want that, 00:57:44.730 --> 00:57:47.040 align:middle line:84% and then I'll tell you why you don't do that. 00:57:47.040 --> 00:57:48.750 align:middle line:84% AUDIENCE: Maybe, then, we would see-- 00:57:48.750 --> 00:57:50.430 align:middle line:84% well, in the first case, when you 00:57:50.430 --> 00:57:54.735 align:middle line:84% were talking about [INAUDIBLE],, I reminded you in the end, 00:57:54.735 --> 00:57:56.880 align:middle line:84% if you can make a connection with the [INAUDIBLE],, 00:57:56.880 --> 00:57:58.590 align:middle line:90% you actually risk overshooting. 00:57:58.590 --> 00:57:58.965 align:middle line:90% BEN OLKEN: Overshooting? 00:57:58.965 --> 00:58:00.220 align:middle line:90% AUDIENCE: Yeah, yeah, yeah. 00:58:00.220 --> 00:58:00.887 align:middle line:90% BEN OLKEN: Yeah. 00:58:00.887 --> 00:58:05.400 align:middle line:84% AUDIENCE: And [INAUDIBLE] to not only avoid that, 00:58:05.400 --> 00:58:09.930 align:middle line:84% but also, actually see to what extent 00:58:09.930 --> 00:58:14.520 align:middle line:84% it's [INAUDIBLE] over the [INAUDIBLE] talking about. 00:58:14.520 --> 00:58:15.720 align:middle line:90% BEN OLKEN: Yeah, so OK. 00:58:15.720 --> 00:58:17.290 align:middle line:84% So Patrick's point, just to be clear, 00:58:17.290 --> 00:58:19.740 align:middle line:84% is, in our example, maybe I should 00:58:19.740 --> 00:58:22.860 align:middle line:84% do multiple shocks in case one of them might overshoot. 00:58:22.860 --> 00:58:26.613 align:middle line:84% So with infinite sample and infinite budget, agree. 00:58:26.613 --> 00:58:28.530 align:middle line:84% The reason that they don't do that in practice 00:58:28.530 --> 00:58:36.420 align:middle line:90% is a power issue, basically. 00:58:36.420 --> 00:58:39.270 align:middle line:84% So if I split my sample, imagine I-- 00:58:39.270 --> 00:58:41.520 align:middle line:84% this is a case, by the way, where your treatment costs 00:58:41.520 --> 00:58:42.540 align:middle line:90% a lot more than your control. 00:58:42.540 --> 00:58:44.080 align:middle line:84% You have to pay for all these jobs. 00:58:44.080 --> 00:58:47.175 align:middle line:84% So the relevant case is not even instead of half-half, 00:58:47.175 --> 00:58:49.050 align:middle line:84% like a third, a third, a third, it's probably 00:58:49.050 --> 00:58:52.110 align:middle line:84% like a quarter, a quarter, control, 00:58:52.110 --> 00:58:54.030 align:middle line:84% is the budget-neutral thing to do. 00:58:54.030 --> 00:58:56.508 align:middle line:90% 00:58:56.508 --> 00:58:58.800 align:middle line:84% Approximately, not quite, because you have fewer people 00:58:58.800 --> 00:59:00.720 align:middle line:84% employed and then a smaller shock group, 00:59:00.720 --> 00:59:02.520 align:middle line:90% but let's call it roughly that. 00:59:02.520 --> 00:59:06.132 align:middle line:84% So then-- actually, no, I take that back. 00:59:06.132 --> 00:59:08.090 align:middle line:84% That could be, actually, first order, actually. 00:59:08.090 --> 00:59:10.240 align:middle line:90% Because you have another-- 00:59:10.240 --> 00:59:12.196 align:middle line:84% actually, you have another question here. 00:59:12.196 --> 00:59:14.690 align:middle line:84% OK, there's two experimental design issues. 00:59:14.690 --> 00:59:18.760 align:middle line:84% The first experimental design issue is, 00:59:18.760 --> 00:59:22.060 align:middle line:84% is it better to have a smaller number of villages 00:59:22.060 --> 00:59:26.410 align:middle line:84% with a big shock, or more villages with a small shock? 00:59:26.410 --> 00:59:27.740 align:middle line:90% So actually, that is relevant. 00:59:27.740 --> 00:59:30.430 align:middle line:84% So you could-- for the same price, 00:59:30.430 --> 00:59:35.920 align:middle line:84% I could have 100 villages with a 24% shock, or 200 treatment 00:59:35.920 --> 00:59:40.990 align:middle line:90% villages with a 12% shock. 00:59:40.990 --> 00:59:43.190 align:middle line:90% This is the same cost. 00:59:43.190 --> 00:59:44.210 align:middle line:90% Which is better? 00:59:44.210 --> 00:59:47.840 align:middle line:84% Forgetting about this issue for a second, just general straight 00:59:47.840 --> 00:59:48.340 align:middle line:90% up-- 00:59:48.340 --> 00:59:50.493 align:middle line:84% like forget about this more complicated issue. 00:59:50.493 --> 00:59:52.660 align:middle line:84% Just straight up experimental design, which of these 00:59:52.660 --> 00:59:54.769 align:middle line:90% do you prefer? 00:59:54.769 --> 00:59:58.000 align:middle line:84% AUDIENCE: The [INAUDIBLE] one seems to have more power, 00:59:58.000 --> 00:59:58.760 align:middle line:90% I think. 00:59:58.760 --> 01:00:01.132 align:middle line:84% BEN OLKEN: Yeah, so does everyone understand why? 01:00:01.132 --> 01:00:02.840 align:middle line:84% Did I go over this already in this class, 01:00:02.840 --> 01:00:05.410 align:middle line:84% or was that in the other class I taught this? 01:00:05.410 --> 01:00:06.400 align:middle line:90% I can't keep track. 01:00:06.400 --> 01:00:08.170 align:middle line:84% Aaron, do you remember which class I went over this issue? 01:00:08.170 --> 01:00:09.190 align:middle line:90% AUDIENCE: [INAUDIBLE]. 01:00:09.190 --> 01:00:10.030 align:middle line:90% BEN OLKEN: OK, someone else? 01:00:10.030 --> 01:00:10.930 align:middle line:90% Relevant to someone. 01:00:10.930 --> 01:00:12.597 align:middle line:84% Someone else, did I explain this already 01:00:12.597 --> 01:00:14.165 align:middle line:84% with the beta and the T-stat thing? 01:00:14.165 --> 01:00:14.665 align:middle line:90% Yes? 01:00:14.665 --> 01:00:15.468 align:middle line:90% OK, great, exactly. 01:00:15.468 --> 01:00:16.760 align:middle line:90% Then you were paying attention. 01:00:16.760 --> 01:00:19.423 align:middle line:90% Good, so exactly. 01:00:19.423 --> 01:00:20.590 align:middle line:90% This one has a higher power. 01:00:20.590 --> 01:00:22.810 align:middle line:90% So the first point is that you-- 01:00:22.810 --> 01:00:27.700 align:middle line:84% the lower shocks, you have less power. 01:00:27.700 --> 01:00:32.390 align:middle line:84% The second point is related to that, 01:00:32.390 --> 01:00:41.830 align:middle line:84% which is if you don't know if you're going to have an effect 01:00:41.830 --> 01:00:46.400 align:middle line:84% or be able to detect it, it is almost always better 01:00:46.400 --> 01:00:49.280 align:middle line:84% to have just one control group and a really big treatment 01:00:49.280 --> 01:00:51.660 align:middle line:84% group, rather than an intermediate effect. 01:00:51.660 --> 01:01:01.040 align:middle line:84% And that's because-- it's like if this is the control, 01:01:01.040 --> 01:01:05.280 align:middle line:84% and this is the treatment, in general, 01:01:05.280 --> 01:01:07.940 align:middle line:84% we think the response of most things is linear-- 01:01:07.940 --> 01:01:17.027 align:middle line:84% like approximately linear-- then we're trying to figure out, 01:01:17.027 --> 01:01:18.610 align:middle line:84% we have this kind of small thing here. 01:01:18.610 --> 01:01:20.860 align:middle line:84% We have, normally, a lot of power on that small thing. 01:01:20.860 --> 01:01:24.525 align:middle line:84% And we'd be trying to test, are we on this linear-- 01:01:24.525 --> 01:01:25.900 align:middle line:84% like we're usually interested in, 01:01:25.900 --> 01:01:27.920 align:middle line:84% are we on this linear curve or not? 01:01:27.920 --> 01:01:36.130 align:middle line:84% And so testing not just like, are you-- 01:01:36.130 --> 01:01:37.840 align:middle line:84% identifying this lower range thing 01:01:37.840 --> 01:01:39.910 align:middle line:84% and testing whether you're on this linear curve 01:01:39.910 --> 01:01:42.070 align:middle line:90% requires a lot of power. 01:01:42.070 --> 01:01:46.240 align:middle line:84% So most of the time, these intermediate outcomes 01:01:46.240 --> 01:01:47.980 align:middle line:84% are very, very difficult to distinguish 01:01:47.980 --> 01:01:51.002 align:middle line:84% from being on the linear, just on the linear curve. 01:01:51.002 --> 01:01:52.710 align:middle line:84% Sorry, the reason for that, I should say, 01:01:52.710 --> 01:01:54.340 align:middle line:84% is like you have less power here, 01:01:54.340 --> 01:01:57.855 align:middle line:84% and you're trying to test that this coefficient is 01:01:57.855 --> 01:01:59.230 align:middle line:84% equal to 1/2 of that coefficient. 01:01:59.230 --> 01:02:02.820 align:middle line:90% It's a very tight test. 01:02:02.820 --> 01:02:05.580 align:middle line:84% Now, so that's the general reason 01:02:05.580 --> 01:02:09.182 align:middle line:84% why, if you're not sure if you have an effect, 01:02:09.182 --> 01:02:10.890 align:middle line:84% you would need a really big sample for it 01:02:10.890 --> 01:02:15.400 align:middle line:84% to make sense to do these 1/2 power intermediate outcomes. 01:02:15.400 --> 01:02:17.163 align:middle line:84% In your case, that's not quite right. 01:02:17.163 --> 01:02:18.580 align:middle line:84% In this case, it's not quite right 01:02:18.580 --> 01:02:20.930 align:middle line:84% because we have this additional hypothesis, 01:02:20.930 --> 01:02:25.420 align:middle line:84% this additional issue of maybe we're overshooting. 01:02:25.420 --> 01:02:27.040 align:middle line:90% But again, the problem is-- 01:02:27.040 --> 01:02:39.010 align:middle line:90% 01:02:39.010 --> 01:02:40.630 align:middle line:84% if you ran that test, you would want 01:02:40.630 --> 01:02:44.575 align:middle line:84% to say, well, suppose you found in that case that-- 01:02:44.575 --> 01:02:49.540 align:middle line:84% the only relevant case is the case where, in the full one, 01:02:49.540 --> 01:02:51.880 align:middle line:84% we find labor supply and labor demand effects, 01:02:51.880 --> 01:02:53.590 align:middle line:90% and in the smaller, we don't. 01:02:53.590 --> 01:02:56.530 align:middle line:90% That's the overshooting case. 01:02:56.530 --> 01:02:58.870 align:middle line:84% But then, you have the other problem of, well, 01:02:58.870 --> 01:03:01.340 align:middle line:84% the treatment was also half the size. 01:03:01.340 --> 01:03:03.958 align:middle line:84% So you also have to-- you'd have to rule out not only-- you're 01:03:03.958 --> 01:03:05.000 align:middle line:90% kind of in the same case. 01:03:05.000 --> 01:03:06.650 align:middle line:90% You have to rule out-- 01:03:06.650 --> 01:03:10.430 align:middle line:84% you have to rule out, not only are these things different-- 01:03:10.430 --> 01:03:12.145 align:middle line:84% like this one's zero and this one's not-- 01:03:12.145 --> 01:03:13.520 align:middle line:84% but actually, you have to also be 01:03:13.520 --> 01:03:15.260 align:middle line:84% able to reject a harder thing, which 01:03:15.260 --> 01:03:18.310 align:middle line:84% is this is not equal to just half of this effect. 01:03:18.310 --> 01:03:19.310 align:middle line:90% You see what I'm saying? 01:03:19.310 --> 01:03:21.830 align:middle line:84% The thing you're testing is not just that this would be 0 01:03:21.830 --> 01:03:23.030 align:middle line:84% and this would be the full thing, 01:03:23.030 --> 01:03:25.160 align:middle line:84% but you also have to reject that this thing is not just half 01:03:25.160 --> 01:03:27.140 align:middle line:84% of that, because that would just be linear. 01:03:27.140 --> 01:03:27.930 align:middle line:90% You see what I'm saying? 01:03:27.930 --> 01:03:29.472 align:middle line:84% So the amount of power that you would 01:03:29.472 --> 01:03:34.072 align:middle line:84% need to do that successfully would be pretty large. 01:03:34.072 --> 01:03:35.495 align:middle line:90% Does that answer your question? 01:03:35.495 --> 01:03:36.120 align:middle line:90% AUDIENCE: Yeah. 01:03:36.120 --> 01:03:38.130 align:middle line:84% BEN OLKEN: Yeah, so I think that basically, I mean, 01:03:38.130 --> 01:03:40.297 align:middle line:84% another way of saying it is I think on these things, 01:03:40.297 --> 01:03:43.435 align:middle line:90% your best bet is-- 01:03:43.435 --> 01:03:45.060 align:middle line:84% even though it's very frustrating, very 01:03:45.060 --> 01:03:48.447 align:middle line:84% nerve-wracking as a researcher, it is often not the case 01:03:48.447 --> 01:03:50.280 align:middle line:84% that the best thing to do is hedge your bets 01:03:50.280 --> 01:03:52.680 align:middle line:84% with two different treatment effect sizes. 01:03:52.680 --> 01:03:55.358 align:middle line:84% And by the way, if you see a lot of the-- 01:03:55.358 --> 01:03:58.545 align:middle line:90% 01:03:58.545 --> 01:04:00.170 align:middle line:84% if you think about how they're running, 01:04:00.170 --> 01:04:02.790 align:middle line:84% how they run clinical trials, if you think about the COVID 01:04:02.790 --> 01:04:05.290 align:middle line:84% vaccine clinical trials-- which are probably clinical trials 01:04:05.290 --> 01:04:08.140 align:middle line:84% everyone is most familiar with, I'm guessing, at the moment-- 01:04:08.140 --> 01:04:09.534 align:middle line:90% that's what they did. 01:04:09.534 --> 01:04:13.720 align:middle line:84% The first trial of, are we going to approve the vaccine or not, 01:04:13.720 --> 01:04:16.330 align:middle line:84% they did a bunch of little small things to check the dosage 01:04:16.330 --> 01:04:20.260 align:middle line:84% and make sure it seemed to be having antibody responses 01:04:20.260 --> 01:04:22.070 align:middle line:84% and not being toxic, or whatever. 01:04:22.070 --> 01:04:26.170 align:middle line:84% And they just took their best shot and picked one protocol 01:04:26.170 --> 01:04:28.525 align:middle line:84% and ran with that for the phase III trial. 01:04:28.525 --> 01:04:32.230 align:middle line:90% 01:04:32.230 --> 01:04:34.742 align:middle line:84% And then, once it worked, and then in subsequent trials, 01:04:34.742 --> 01:04:36.700 align:middle line:84% then they're starting to test different things. 01:04:36.700 --> 01:04:38.700 align:middle line:84% And then they have a much better sense of power. 01:04:38.700 --> 01:04:40.540 align:middle line:84% And also, the thing works, so they cam-- 01:04:40.540 --> 01:04:42.290 align:middle line:84% but the very first trial of, is this thing 01:04:42.290 --> 01:04:44.790 align:middle line:84% going to work, they were like, we don't know what the effect 01:04:44.790 --> 01:04:46.210 align:middle line:84% size is going to be, so we're just 01:04:46.210 --> 01:04:48.610 align:middle line:84% going to max out treatment versus control 01:04:48.610 --> 01:04:50.950 align:middle line:84% and just take our best shot based 01:04:50.950 --> 01:04:52.930 align:middle line:84% on the preliminary small-scale studies of what 01:04:52.930 --> 01:04:53.840 align:middle line:90% we think is going to work. 01:04:53.840 --> 01:04:56.340 align:middle line:84% And that's what we're going to do for the big trial and not, 01:04:56.340 --> 01:04:59.800 align:middle line:84% in the very first trial, go for multiple different strategies. 01:04:59.800 --> 01:05:01.705 align:middle line:84% As I said, once it works, then you actually 01:05:01.705 --> 01:05:02.830 align:middle line:90% have much more information. 01:05:02.830 --> 01:05:03.850 align:middle line:84% You know how big the response is. 01:05:03.850 --> 01:05:05.440 align:middle line:84% You can plot these things much more carefully. 01:05:05.440 --> 01:05:06.730 align:middle line:90% You have much tighter priors. 01:05:06.730 --> 01:05:08.320 align:middle line:84% But for that first efficacy trial, 01:05:08.320 --> 01:05:10.933 align:middle line:84% it was just like treatment versus control. 01:05:10.933 --> 01:05:11.850 align:middle line:90% AUDIENCE: [INAUDIBLE]? 01:05:11.850 --> 01:05:12.240 align:middle line:90% BEN OLKEN: Yeah. 01:05:12.240 --> 01:05:13.365 align:middle line:90% AUDIENCE: If you're doing-- 01:05:13.365 --> 01:05:15.660 align:middle line:90% 01:05:15.660 --> 01:05:17.160 align:middle line:84% I guess if you go for one big group, 01:05:17.160 --> 01:05:21.990 align:middle line:84% you're not able to ask questions about dosage for the trials. 01:05:21.990 --> 01:05:24.812 align:middle line:84% But you can ask questions about other sorts of-- 01:05:24.812 --> 01:05:26.770 align:middle line:84% I mean, I guess dosage is different [INAUDIBLE] 01:05:26.770 --> 01:05:27.270 align:middle line:90% in clinic. 01:05:27.270 --> 01:05:30.688 align:middle line:84% But my point is, you can still do more than you are. 01:05:30.688 --> 01:05:32.730 align:middle line:84% I'm trying to think like what subsequent analysis 01:05:32.730 --> 01:05:38.520 align:middle line:84% can you do beyond the simple difference of means? 01:05:38.520 --> 01:05:40.820 align:middle line:90% BEN OLKEN: What do you mean? 01:05:40.820 --> 01:05:46.775 align:middle line:84% AUDIENCE: Like-- You lose the ability to do doses, 01:05:46.775 --> 01:05:47.557 align:middle line:90% [INAUDIBLE]. 01:05:47.557 --> 01:05:48.390 align:middle line:90% BEN OLKEN: Yes, yes. 01:05:48.390 --> 01:05:50.432 align:middle line:84% You cannot-- exactly, you can't do dose-response. 01:05:50.432 --> 01:05:51.932 align:middle line:84% AUDIENCE: In a sentence, like what's 01:05:51.932 --> 01:05:53.038 align:middle line:90% your thought [INAUDIBLE]? 01:05:53.038 --> 01:05:57.520 align:middle line:90% 01:05:57.520 --> 01:05:59.380 align:middle line:84% BEN OLKEN: I mean, if you're trading off-- 01:05:59.380 --> 01:06:01.797 align:middle line:84% if you don't do dose-response, you can't do dose-response. 01:06:01.797 --> 01:06:02.750 align:middle line:90% That's what you lose. 01:06:02.750 --> 01:06:04.900 align:middle line:84% I think that that's the downside. 01:06:04.900 --> 01:06:05.983 align:middle line:90% No, so what's the gain? 01:06:05.983 --> 01:06:07.900 align:middle line:84% The gain is you're much more likely to pick up 01:06:07.900 --> 01:06:10.060 align:middle line:90% an effect by, like, a lot. 01:06:10.060 --> 01:06:11.298 align:middle line:90% That's the main. 01:06:11.298 --> 01:06:13.840 align:middle line:84% If you're just trying to say, does this thing have an effect, 01:06:13.840 --> 01:06:16.330 align:middle line:84% and the first-order estimate of what 01:06:16.330 --> 01:06:20.685 align:middle line:84% is the slope of that thing, the first-order things 01:06:20.685 --> 01:06:22.060 align:middle line:84% are going to get a lot more power 01:06:22.060 --> 01:06:25.265 align:middle line:84% on identifying what's going on and rejecting the null of 0. 01:06:25.265 --> 01:06:26.890 align:middle line:84% AUDIENCE: Wouldn't you do heterogeneity 01:06:26.890 --> 01:06:30.050 align:middle line:84% based on people that you think [INAUDIBLE] might 01:06:30.050 --> 01:06:31.310 align:middle line:90% need a larger or smaller dose? 01:06:31.310 --> 01:06:32.840 align:middle line:84% BEN OLKEN: You could if you-- yeah, 01:06:32.840 --> 01:06:34.980 align:middle line:84% so if you think of the clinical trial example, 01:06:34.980 --> 01:06:36.620 align:middle line:90% if you thought that if you-- 01:06:36.620 --> 01:06:38.703 align:middle line:84% maybe you thought dose was proportional to weight. 01:06:38.703 --> 01:06:40.828 align:middle line:84% For example, you could do that if you [INAUDIBLE].. 01:06:40.828 --> 01:06:42.630 align:middle line:84% AUDIENCE: So that's OK [INAUDIBLE].. 01:06:42.630 --> 01:06:45.090 align:middle line:84% BEN OLKEN: But it's not randomized heterogeneity. 01:06:45.090 --> 01:06:46.632 align:middle line:84% So it's correlated with other things. 01:06:46.632 --> 01:06:48.840 align:middle line:84% Maybe like people who are, whatever, 01:06:48.840 --> 01:06:51.900 align:middle line:84% have different immune system, or whoever the equivalent is here. 01:06:51.900 --> 01:06:54.420 align:middle line:90% 01:06:54.420 --> 01:06:58.258 align:middle line:90% OK, all right. 01:06:58.258 --> 01:06:59.300 align:middle line:90% OK, so what do they find? 01:06:59.300 --> 01:07:02.410 align:middle line:84% So they find that basically, in the peak season, 01:07:02.410 --> 01:07:03.580 align:middle line:90% wages increase by 5%. 01:07:03.580 --> 01:07:06.790 align:middle line:90% Employment declines by 21%. 01:07:06.790 --> 01:07:12.140 align:middle line:84% And in the slack season, basically, wages are unchanged. 01:07:12.140 --> 01:07:15.520 align:middle line:84% So the peak season looks like a competitive labor 01:07:15.520 --> 01:07:18.380 align:middle line:84% market with elasticity of labor demand of around minus 4. 01:07:18.380 --> 01:07:20.960 align:middle line:90% 01:07:20.960 --> 01:07:27.360 align:middle line:84% And the slack season looks like you have rationing. 01:07:27.360 --> 01:07:29.560 align:middle line:90% OK, so both of those-- 01:07:29.560 --> 01:07:31.560 align:middle line:84% in some sense, that paper and the previous paper 01:07:31.560 --> 01:07:34.410 align:middle line:84% go together to basically say, look, 01:07:34.410 --> 01:07:37.650 align:middle line:84% it seems like we have something not clearing 01:07:37.650 --> 01:07:38.640 align:middle line:90% in the labor market. 01:07:38.640 --> 01:07:41.053 align:middle line:90% The [INAUDIBLE] was saying that. 01:07:41.053 --> 01:07:42.720 align:middle line:84% That experimental paper was saying that. 01:07:42.720 --> 01:07:44.345 align:middle line:84% There's something in these global labor 01:07:44.345 --> 01:07:47.360 align:middle line:84% markets which is preventing labor market clearing. 01:07:47.360 --> 01:07:48.740 align:middle line:90% So the question is, why? 01:07:48.740 --> 01:07:50.420 align:middle line:90% What's going on? 01:07:50.420 --> 01:07:53.010 align:middle line:90% And why is this happening? 01:07:53.010 --> 01:07:54.740 align:middle line:84% So there's a couple of recent papers 01:07:54.740 --> 01:07:56.430 align:middle line:84% who have started to investigate this. 01:07:56.430 --> 01:08:02.580 align:middle line:84% So one is another paper by a similar set 01:08:02.580 --> 01:08:07.340 align:middle line:84% of co-authors which runs an experiment to get at this. 01:08:07.340 --> 01:08:08.340 align:middle line:90% And here's what they do. 01:08:08.340 --> 01:08:11.100 align:middle line:90% 01:08:11.100 --> 01:08:14.450 align:middle line:84% They go to Indian villages, and they offer spot jobs, 01:08:14.450 --> 01:08:19.140 align:middle line:84% They vary the wage of the jobs, and they vary whether the wage 01:08:19.140 --> 01:08:21.520 align:middle line:90% offer is observable or not. 01:08:21.520 --> 01:08:23.973 align:middle line:84% And the way they do this is pretty subtle. 01:08:23.973 --> 01:08:25.890 align:middle line:84% They basically do it in the context of survey. 01:08:25.890 --> 01:08:27.890 align:middle line:84% They do a baseline survey, and they say, by the, 01:08:27.890 --> 01:08:29.603 align:middle line:90% way we're offering this job. 01:08:29.603 --> 01:08:30.270 align:middle line:90% Here's the wage. 01:08:30.270 --> 01:08:32.220 align:middle line:90% Would you like to do it? 01:08:32.220 --> 01:08:35.279 align:middle line:84% And in some of them, they make that offer 01:08:35.279 --> 01:08:37.859 align:middle line:84% on the front porch of the person's house. 01:08:37.859 --> 01:08:39.997 align:middle line:84% So people who are-- and if any of you 01:08:39.997 --> 01:08:42.330 align:middle line:84% have ever done a survey in a developing country context, 01:08:42.330 --> 01:08:44.189 align:middle line:84% it's often pretty common for other people 01:08:44.189 --> 01:08:46.439 align:middle line:84% to crowd around and listen in to what's going on. 01:08:46.439 --> 01:08:47.700 align:middle line:90% It's pretty exciting. 01:08:47.700 --> 01:08:49.950 align:middle line:84% Some person's here doing a survey. 01:08:49.950 --> 01:08:50.700 align:middle line:90% Let's listen in. 01:08:50.700 --> 01:08:53.430 align:middle line:84% So they do that on the front porch. 01:08:53.430 --> 01:08:55.407 align:middle line:84% Or they say, let's please do the same thing, 01:08:55.407 --> 01:08:57.990 align:middle line:84% but we're going to do it inside your house so it's in private. 01:08:57.990 --> 01:08:59.430 align:middle line:84% So it's exactly the same, but just 01:08:59.430 --> 01:09:01.920 align:middle line:84% varying whether it's on your front porch or in private. 01:09:01.920 --> 01:09:06.920 align:middle line:90% 01:09:06.920 --> 01:09:09.845 align:middle line:84% So those are the three treatments 01:09:09.845 --> 01:09:11.845 align:middle line:84% And so what would you what you expect from this? 01:09:11.845 --> 01:09:14.362 align:middle line:90% 01:09:14.362 --> 01:09:16.529 align:middle line:84% AUDIENCE: You're more likely to accept a lower grade 01:09:16.529 --> 01:09:17.237 align:middle line:90% in private than-- 01:09:17.237 --> 01:09:18.060 align:middle line:90% BEN OLKEN: Why? 01:09:18.060 --> 01:09:19.140 align:middle line:90% AUDIENCE: Pride. 01:09:19.140 --> 01:09:20.430 align:middle line:90% Pride, essentially. 01:09:20.430 --> 01:09:23.580 align:middle line:84% BEN OLKEN: OK, so any other theories of why? 01:09:23.580 --> 01:09:26.750 align:middle line:90% 01:09:26.750 --> 01:09:27.380 align:middle line:90% Yeah? 01:09:27.380 --> 01:09:30.240 align:middle line:90% AUDIENCE: I guess there's also-- 01:09:30.240 --> 01:09:33.674 align:middle line:90% if there's a-- 01:09:33.674 --> 01:09:37.220 align:middle line:84% I guess the analogy I have in my mind is like a scab, 01:09:37.220 --> 01:09:40.920 align:middle line:84% basically accepting a lower wage when people are-- 01:09:40.920 --> 01:09:43.306 align:middle line:84% if there's this agreed upon [INAUDIBLE] 01:09:43.306 --> 01:09:47.024 align:middle line:84% we're not going to work for [INAUDIBLE].. 01:09:47.024 --> 01:09:49.149 align:middle line:84% BEN OLKEN: Right, and in fact, the very first draft 01:09:49.149 --> 01:09:50.590 align:middle line:84% of this paper was titled "Scabs." 01:09:50.590 --> 01:09:52.807 align:middle line:84% And then they changed title, in part 01:09:52.807 --> 01:09:54.640 align:middle line:84% to acknowledge that there were other things, 01:09:54.640 --> 01:09:55.450 align:middle line:90% I think, going on. 01:09:55.450 --> 01:09:59.320 align:middle line:84% But yeah, so one theory is like this external social norm 01:09:59.320 --> 01:09:59.980 align:middle line:90% theory-- 01:09:59.980 --> 01:10:03.890 align:middle line:84% that basically, there's a lot of peer pressure 01:10:03.890 --> 01:10:07.060 align:middle line:84% that's going on to prevent, to enforce these higher wages. 01:10:07.060 --> 01:10:09.310 align:middle line:84% And so any individual is going to get social sanctions 01:10:09.310 --> 01:10:10.360 align:middle line:90% if they accept the lower wage. 01:10:10.360 --> 01:10:13.060 align:middle line:84% And the other is your theory, is like this is an internal thing 01:10:13.060 --> 01:10:15.670 align:middle line:90% and I'm not going to do it. 01:10:15.670 --> 01:10:18.055 align:middle line:84% But the internal one would have to be only my pride 01:10:18.055 --> 01:10:19.180 align:middle line:90% vis a vis the other people. 01:10:19.180 --> 01:10:20.305 align:middle line:90% It's not my internal pride. 01:10:20.305 --> 01:10:21.838 align:middle line:84% It's that people are watching me. 01:10:21.838 --> 01:10:23.380 align:middle line:84% AUDIENCE: Yeah, I think this is maybe 01:10:23.380 --> 01:10:25.810 align:middle line:84% a little bit of articulation of the dynamic that I had. 01:10:25.810 --> 01:10:26.560 align:middle line:90% BEN OLKEN: What, sorry? 01:10:26.560 --> 01:10:28.480 align:middle line:84% AUDIENCE: This is maybe a slightly better articulation 01:10:28.480 --> 01:10:29.897 align:middle line:84% of the dynamic that I had in mind. 01:10:29.897 --> 01:10:30.770 align:middle line:90% BEN OLKEN: Got it. 01:10:30.770 --> 01:10:32.187 align:middle line:84% So there's something going, right? 01:10:32.187 --> 01:10:38.200 align:middle line:84% And so by the way, can you imagine, 01:10:38.200 --> 01:10:40.940 align:middle line:84% are there cases where it would be optimal, maybe, 01:10:40.940 --> 01:10:43.325 align:middle line:84% for the village to have these norms? 01:10:43.325 --> 01:10:48.590 align:middle line:90% 01:10:48.590 --> 01:10:51.040 align:middle line:84% AUDIENCE: [INAUDIBLE] a monopsony, then [INAUDIBLE]?? 01:10:51.040 --> 01:10:52.000 align:middle line:90% BEN OLKEN: Yeah, exactly, right? 01:10:52.000 --> 01:10:53.650 align:middle line:84% So if there's a monopsony power over the local labor market 01:10:53.650 --> 01:10:54.790 align:middle line:84% power, it's like exactly, it's kind 01:10:54.790 --> 01:10:56.170 align:middle line:90% of bargaining with the firm. 01:10:56.170 --> 01:10:58.170 align:middle line:84% And/or in general, if there's any kind of rents, 01:10:58.170 --> 01:10:59.920 align:middle line:84% we might want to extract from those rents, 01:10:59.920 --> 01:11:02.020 align:middle line:84% and so we want to keep the wage artificially high. 01:11:02.020 --> 01:11:03.850 align:middle line:84% And that might mean, whenever we keep 01:11:03.850 --> 01:11:05.770 align:middle line:84% the wage artificially high, there's 01:11:05.770 --> 01:11:07.570 align:middle line:90% a little bit of surplus. 01:11:07.570 --> 01:11:10.750 align:middle line:84% Individuals might want to deviate. 01:11:10.750 --> 01:11:14.050 align:middle line:84% But we're all trying to keep the wage high to extract rents 01:11:14.050 --> 01:11:14.740 align:middle line:90% together. 01:11:14.740 --> 01:11:17.590 align:middle line:84% So it's kind of a collective action model. 01:11:17.590 --> 01:11:19.750 align:middle line:84% OK, and here's the result. It's super simple. 01:11:19.750 --> 01:11:24.160 align:middle line:90% 01:11:24.160 --> 01:11:27.260 align:middle line:84% The result is-- this is the main result. 01:11:27.260 --> 01:11:31.840 align:middle line:84% So this is probability of taking up the wage at the going market 01:11:31.840 --> 01:11:33.430 align:middle line:90% wage. 01:11:33.430 --> 01:11:37.730 align:middle line:84% At 10% less the going market wage, when in public-- 01:11:37.730 --> 01:11:39.320 align:middle line:90% I'm sorry, in Private-- 01:11:39.320 --> 01:11:43.130 align:middle line:84% and 10% less than the market wage in public. 01:11:43.130 --> 01:11:44.060 align:middle line:90% So it's pretty stark. 01:11:44.060 --> 01:11:47.300 align:middle line:84% So basically, almost nobody is willing to-- 01:11:47.300 --> 01:11:50.120 align:middle line:84% and I'm sorry, this is in this the high unemployment periods 01:11:50.120 --> 01:11:51.333 align:middle line:90% and low unemployment periods. 01:11:51.333 --> 01:11:53.750 align:middle line:84% In the high unemployment period, basically, lots of people 01:11:53.750 --> 01:11:55.333 align:middle line:84% want this job, because there's a lot-- 01:11:55.333 --> 01:11:58.010 align:middle line:84% that thing, the labor market, is really-- 01:11:58.010 --> 01:11:59.360 align:middle line:90% that gap is kind of large. 01:11:59.360 --> 01:12:02.310 align:middle line:84% In the low unemployment periods, the gap is kind of small. 01:12:02.310 --> 01:12:04.760 align:middle line:84% And so maybe, actually, we don't care that much. 01:12:04.760 --> 01:12:06.770 align:middle line:84% So in these periods over here, that's 01:12:06.770 --> 01:12:10.410 align:middle line:84% the one consistent with what [INAUDIBLE] was suggesting, 01:12:10.410 --> 01:12:13.700 align:middle line:84% which is basically we're in a market 01:12:13.700 --> 01:12:15.350 align:middle line:90% where the wage is too high. 01:12:15.350 --> 01:12:19.010 align:middle line:84% We'd all like to get jobs, so there's some failure 01:12:19.010 --> 01:12:21.763 align:middle line:90% of market clearing. 01:12:21.763 --> 01:12:23.930 align:middle line:84% And so people are like, if you can do it in private, 01:12:23.930 --> 01:12:25.400 align:middle line:84% you're happy to take it, but you don't 01:12:25.400 --> 01:12:27.317 align:middle line:84% want to do it in public because you're worried 01:12:27.317 --> 01:12:30.450 align:middle line:90% about retaliation or whatever. 01:12:30.450 --> 01:12:32.600 align:middle line:84% And in fact, one thing that's kind of nice is-- 01:12:32.600 --> 01:12:33.590 align:middle line:84% this is too small for you to read, 01:12:33.590 --> 01:12:34.965 align:middle line:84% but you can see in your handout-- 01:12:34.965 --> 01:12:39.205 align:middle line:84% they ask people what's going on, and there are various-- 01:12:39.205 --> 01:12:42.540 align:middle line:90% 01:12:42.540 --> 01:12:44.970 align:middle line:84% what you would do, sorry, to a worker 01:12:44.970 --> 01:12:48.570 align:middle line:84% who took one of these jobs or whatever? 01:12:48.570 --> 01:12:50.310 align:middle line:84% You would try to convince them not to. 01:12:50.310 --> 01:12:52.560 align:middle line:84% You would impose social sanctions, et cetera, 01:12:52.560 --> 01:12:53.160 align:middle line:90% et cetera. 01:12:53.160 --> 01:12:55.320 align:middle line:84% You'd try to exclude them from other labor market 01:12:55.320 --> 01:12:58.950 align:middle line:84% opportunities, et cetera, or what? 01:12:58.950 --> 01:13:05.640 align:middle line:84% So in general, it looks like people-- 01:13:05.640 --> 01:13:07.120 align:middle line:84% there are these public punishments 01:13:07.120 --> 01:13:10.770 align:middle line:90% for breaking these wage norms. 01:13:10.770 --> 01:13:14.760 align:middle line:84% And so that's the beginning of an explanation 01:13:14.760 --> 01:13:16.740 align:middle line:84% for how we might be getting these labor market 01:13:16.740 --> 01:13:18.750 align:middle line:90% failures of clearing. 01:13:18.750 --> 01:13:22.170 align:middle line:84% We have these prevailing wage norms. 01:13:22.170 --> 01:13:25.140 align:middle line:84% Those wages move around, and then they 01:13:25.140 --> 01:13:28.260 align:middle line:84% stay high, potentially as a way of getting some rents. 01:13:28.260 --> 01:13:30.120 align:middle line:90% Is it efficient? 01:13:30.120 --> 01:13:31.632 align:middle line:84% It's actually-- to that question, 01:13:31.632 --> 01:13:33.090 align:middle line:84% I'm not sure we have a clear answer 01:13:33.090 --> 01:13:34.230 align:middle line:84% from all these papers of whether or not 01:13:34.230 --> 01:13:37.120 align:middle line:84% this is allocatively efficient for these villagers or not, 01:13:37.120 --> 01:13:40.020 align:middle line:84% whether the additional rents they're getting 01:13:40.020 --> 01:13:42.923 align:middle line:90% exceed the unemployment. 01:13:42.923 --> 01:13:45.090 align:middle line:84% You may be able to calculate that from these papers. 01:13:45.090 --> 01:13:46.840 align:middle line:84% I don't remember them actually doing that, 01:13:46.840 --> 01:13:49.270 align:middle line:84% but that's something to think about as well. 01:13:49.270 --> 01:13:51.580 align:middle line:84% So it looks like, actually-- so stepping way back, 01:13:51.580 --> 01:13:53.260 align:middle line:84% it looks like these labor markets are actually-- 01:13:53.260 --> 01:13:54.718 align:middle line:84% what's going on is kind of nuanced, 01:13:54.718 --> 01:13:56.890 align:middle line:84% having these failures of labor market clearing. 01:13:56.890 --> 01:13:58.390 align:middle line:84% You have these social sanctions that 01:13:58.390 --> 01:14:03.080 align:middle line:90% are potentially impeding that. 01:14:03.080 --> 01:14:05.200 align:middle line:90% OK, questions? 01:14:05.200 --> 01:14:08.240 align:middle line:90% 01:14:08.240 --> 01:14:10.420 align:middle line:84% So the final set of topics on labor supply 01:14:10.420 --> 01:14:13.030 align:middle line:90% that I want to talk about are-- 01:14:13.030 --> 01:14:14.530 align:middle line:84% I won't get to finish them, but I'll 01:14:14.530 --> 01:14:18.070 align:middle line:84% start them-- is some papers about, I would say, 01:14:18.070 --> 01:14:21.380 align:middle line:84% poverty and behavioral issues in labor supply. 01:14:21.380 --> 01:14:31.800 align:middle line:84% So which is to say, so one issue that we talked about in general 01:14:31.800 --> 01:14:34.780 align:middle line:84% was having these labor market frictions. 01:14:34.780 --> 01:14:36.737 align:middle line:84% There may be other reasons, there 01:14:36.737 --> 01:14:39.070 align:middle line:84% may be other things that are affecting these rural labor 01:14:39.070 --> 01:14:40.930 align:middle line:84% markets due to the fact that these people are really poor. 01:14:40.930 --> 01:14:43.250 align:middle line:84% And so how can we think about those issues? 01:14:43.250 --> 01:14:50.170 align:middle line:84% And so I want to explore four different related issues which 01:14:50.170 --> 01:14:51.430 align:middle line:90% will take us mostly the next-- 01:14:51.430 --> 01:14:52.970 align:middle line:84% I will do this one today, and then the rest of them 01:14:52.970 --> 01:14:53.762 align:middle line:90% we'll do next time. 01:14:53.762 --> 01:14:58.725 align:middle line:84% So the first is like, as these question are, 01:14:58.725 --> 01:15:00.900 align:middle line:84% how does poverty affect our labor supply decisions? 01:15:00.900 --> 01:15:03.750 align:middle line:84% One argument is that basic consumption needs basically 01:15:03.750 --> 01:15:05.580 align:middle line:84% change the elasticity of labor supply, 01:15:05.580 --> 01:15:07.910 align:middle line:84% and that can have aggregate implications. 01:15:07.910 --> 01:15:10.040 align:middle line:84% A second is a behavioral channel where 01:15:10.040 --> 01:15:14.000 align:middle line:84% poverty and needing to think about how you're 01:15:14.000 --> 01:15:17.690 align:middle line:84% going to meet your daily poverty needs change your productivity 01:15:17.690 --> 01:15:18.727 align:middle line:90% directly. 01:15:18.727 --> 01:15:20.810 align:middle line:84% We'll talk about identity and labor supply issues. 01:15:20.810 --> 01:15:22.768 align:middle line:84% And then, actually, I'll talk about a new paper 01:15:22.768 --> 01:15:25.100 align:middle line:84% on the cognitive benefits of work. 01:15:25.100 --> 01:15:30.500 align:middle line:84% OK, so the first paper I want to mention about poverty and labor 01:15:30.500 --> 01:15:32.810 align:middle line:84% supply is a paper by [INAUDIBLE] Jayachandran, 01:15:32.810 --> 01:15:34.890 align:middle line:84% and the idea here is very simple. 01:15:34.890 --> 01:15:37.773 align:middle line:84% Her idea is that basically, workers 01:15:37.773 --> 01:15:39.065 align:middle line:90% have minimum consumption needs. 01:15:39.065 --> 01:15:41.630 align:middle line:90% 01:15:41.630 --> 01:15:43.070 align:middle line:84% And if you're really poor, or you 01:15:43.070 --> 01:15:46.100 align:middle line:84% have inability to smooth shocks across periods, 01:15:46.100 --> 01:15:48.778 align:middle line:84% or to borrow, or to do anything to smooth, 01:15:48.778 --> 01:15:50.945 align:middle line:84% that's going to make your labor supply less elastic. 01:15:50.945 --> 01:15:54.005 align:middle line:90% 01:15:54.005 --> 01:15:56.380 align:middle line:84% And to be less elastic, if there's less access to credit, 01:15:56.380 --> 01:16:00.370 align:middle line:84% or less ability to migrate, or whatever, and less elastic 01:16:00.370 --> 01:16:05.820 align:middle line:84% wages are going to mean that there's a pecuniary externality 01:16:05.820 --> 01:16:08.380 align:middle line:90% from you to other people. 01:16:08.380 --> 01:16:11.550 align:middle line:84% So in particular, if we contrast a labor 01:16:11.550 --> 01:16:17.730 align:middle line:84% market where labor supply is elastic and looks like this-- 01:16:17.730 --> 01:16:22.350 align:middle line:84% that's elastic-- to one that looks like this, labor supply 01:16:22.350 --> 01:16:28.590 align:middle line:84% inelastic, imagine what happens when we have a negative labor 01:16:28.590 --> 01:16:30.530 align:middle line:90% supply shock. 01:16:30.530 --> 01:16:33.210 align:middle line:84% When we have a negative labor supply shock, 01:16:33.210 --> 01:16:35.580 align:middle line:84% if labor supply is inelastic, the wage 01:16:35.580 --> 01:16:39.820 align:middle line:84% is going to fall more than if it's elastic. 01:16:39.820 --> 01:16:43.240 align:middle line:84% And what that means is that there's a negative 01:16:43.240 --> 01:16:45.172 align:middle line:84% externality-- a pecuniary externality, 01:16:45.172 --> 01:16:47.380 align:middle line:84% which is an externality that happens through prices-- 01:16:47.380 --> 01:16:50.090 align:middle line:84% of having a lot of poor people around you. 01:16:50.090 --> 01:16:53.020 align:middle line:84% So even if we hold my own ability 01:16:53.020 --> 01:16:55.720 align:middle line:84% to smooth shocks constant, if I'm 01:16:55.720 --> 01:16:58.660 align:middle line:84% surrounded by people who are unable to smooth shocks, 01:16:58.660 --> 01:17:00.458 align:middle line:84% they're going to be more inelastic, 01:17:00.458 --> 01:17:02.500 align:middle line:84% and that means the aggregate wage that I'm facing 01:17:02.500 --> 01:17:04.720 align:middle line:84% is going to fall more in these negative periods. 01:17:04.720 --> 01:17:07.230 align:middle line:90% 01:17:07.230 --> 01:17:09.672 align:middle line:90% Is that clear? 01:17:09.672 --> 01:17:11.250 align:middle line:84% OK, actually, I'm out of time now. 01:17:11.250 --> 01:17:13.595 align:middle line:84% So I'll pick up by just starting with the empirics 01:17:13.595 --> 01:17:14.970 align:middle line:84% of this next time, and then we'll 01:17:14.970 --> 01:17:16.710 align:middle line:84% go through these other papers on labor supply. 01:17:16.710 --> 01:17:17.850 align:middle line:84% And I'm going to try, because I'm 01:17:17.850 --> 01:17:19.975 align:middle line:84% running a little bit behind, to finish up the labor 01:17:19.975 --> 01:17:23.040 align:middle line:84% supply-- sorry, the labor lecture next class so 01:17:23.040 --> 01:17:26.330 align:middle line:84% that we can have spend next week on credit. 01:17:26.330 --> 01:17:39.000 align:middle line:90%