Sunday 10 February 2019

Junk Social Science from Maulik Jagnani

China is a huge country- broader than it is long. Sunrise in its Easternmost city is at least 4 hours ahead of Dawn in its far West. Yet, it has just one Time Zone.

By contrast India is longer than it is broad and at much lower latitudes. Indian Standard Time is a half hour ahead of Pakistan Standard Time and a half hour behind Bangladesh Standard Time. Nepal is a quarter of an hour ahead.

Most of the heavily populated, religiously homogeneous portions of India are within a half hour of the Indian meridian.

Thus, no Indian- apart from some in the far North East- is particularly bothered by the fact that India has only one Time zone.

Or so I thought, till I read on the Marginal Revolution blog, of an
' an original and surprising paper, by Maulik Jagnani, who argues that India’s single time zone reduces the quality of sleep, especially of poor children and this reduces the quality of their education.
Poor Indian children, like this poor Indian middle aged man, are not slaves to the clock- nor, indeed, till quite recently, had much access to accurate time-keeping.

India is very different from America or Europe. It is not the case that there is an age-related convention regarding bed-time.

I recall finding the very concept puzzling. It was explained to me that in the West there was something called a 'watershed' after which it was not safe for kids to watch TV because everybody got naked and started having sex while shooting each other because Capitalism is very evil and seeks to inculcate Consumerism and Sexual Perversion and ubiquitous Gun Crime.

 In Gandhian India, kids, might read, after dark- if there was light- but she went to sleep in the same room, often the same bed, as her younger siblings. Even now, very few Indians are affluent enough to give their kids their own rooms and TVs and Video Games. Even in such families, the pressure to succeed academically is so intense that, though sleep may be traded off for academic study, no pernicious effect arising from 'India's single time zone' could operate.
Why does a nominal change impact real variables? The school day starts at more or less the same clock-hour everywhere in India but children go to bed later in places where the sun sets later.
Schools don't start at the same clock-hour in India. Traditionally, for sociological reasons, elite private schools started earlier and finished earlier than the fee paying Government Schools. Government run Free Schools start later still but are often of poor quality. Cheap private schools must accommodate the schedules of the parents and compete on that basis.

In the Tropics, there is no correlation between time of sunset and going to bed save by reason of excessive heat. For the lower half of the country, there is little seasonal variation in day length- less than an hour on average. Even in the far North it is less than two hours. Thus, there is little variance in sunset time. Sleep may, however, be affected by extreme heat or humidity.

Generally speaking, kids get sleepy when they are tired. Their body clocks can't be affected by Indian Standard Time because there is no mechanism by which any influence, malign or otherwise could be transmitted. Culturally, the tradition is for both parents and kids to sleep early and wake early. But there was no canonical 'bed-time' expressed by the hands of a clock or watch.

 Thus, children in the west get less sleep than children in the east and this shows up in their education levels and later even in their wages!
This is sheer nonsense. The West does better in education levels and wages. The East lags behind.

India has fallen behind China in STEM subjects. How did this happen? One reason is that even free, Government run, Chinese schools teach for more hours and to a more exacting standard than their Indian counterparts. Furthermore, the School Week ends on Saturday.

According to an OECD report-

 on average, pupils in Shanghai aged 12-14 spend 9.8 hours on learning in the classroom, and 3 hours finishing their homework each day, averaging 13.8 hours per week. This is far more than the OECD average of 1.2 hours per day. More than 65% of pupils get up between 6am to 6:30am and go to bed between 10pm to 11:30pm.
So, kids are getting 7 to 8 hours of sleep. My guess is that they have a lunch time nap or catch 40 winks when convenient. This does not damage them cognitively nor does it reduce their future earnings. The opposite is the case.

 In practice, Indian parents who want their kids to do well send them to coaching classes or hire private tutors. The trouble is, the Chinese also do this as did the Japanese before them. No doubt, this means kids have less time to play and that they often wake up groggy. Still, if not all of them, at least some of them, go on to do very well. Human beings are plastic in this respect.

If having a single time zone affects education badly, then China must be a real mess. Kids in Urumqi would getting 4 hours less sleep than kids in Shanghai. They would have developed all sorts of physiological and psychiatric problems.

By contrast, no such effect would be visible in India because half an hour one way or the other is inconsequential.

Jagnani's paper says-

I find that later sunset causes school-age children to begin sleep later, but does not affect wake-up times. An hour (approximately two standard deviation) delay in sunset time reduces children’s sleep by 30 minutes. I also show that later sunset reduces students’ time spent on homework or studying, and time spent on formal and informal work by child laborers,while increasing time spent on indoor leisure for all children. This result is consistent with a model where sleep is productivity-enhancing and increases the marginal returns of study effort for students and work effort for child laborers.
The second part of the paper examines the consequent lifetime impacts of later sunset on stock indicators of children’s academic outcomes. I use nationally-representative data from the 2015 India Demographic and Health Survey (DHS) to estimate how children’s education outcomes co-vary with annual average sunset time across eastern and western locations within a district. I find that an hour (approximately two standard deviation)delay in annual average sunset time reduces years of education by 0.8 years, and children in geographic locations with later sunset are less likely to complete primary and middle school.
Jagnani is asli desi- his first degree is from Mumbai- but what he says makes no sense. Private Schools start and end earlier than Government Schools which however open well after dawn and close well before dusk. Poor kids, having less access to TV and Computers and so on tend to go to bed soon after darkness falls. The seasonal variance in dawn and dusk times is not very great- scarcely an hour- because the Tropic of Cancer runs through the middle of India.

I suppose Educational outcomes could vary with 'average sunset times' in a sufficiently broad country if State and District level authorities were powerless to vary School opening times. This is not the case in India. No State is expansive enough to require 2 Time zones. Education in India has always been a State subject, though it was put on the concurrent list during Indira's brief flirtation with Fascism.

How does Jagnani justify his ludicrous conclusion? The answer is he takes data from
the 1998-99 Indian Time Use Survey (ITUS)
to evaluate
the effect of later sunset on children’s time use. ITUS provides 24-hour time use data, collected with less than a 24-hour recall lapse, allowing me to assign each observation a district-date sunset time.
That survey was the first and last of its kind. Being based on interviews, it was useless. People just made up numbers. This was before mobiles became ubiquitous. Indeed, back then, few poorer women had watches or, indeed, had much need for them. Another survey, from about the same period, showed that only 14 percent of the poor owned a watch or clock. Thus, at least two thirds of the population could not give accurate self-reported input for ITUS because they simply didn't have any way to measure time. Jagnani nevertheless uses the ITUS data to establish a link between later sunset time- later by as little as 5 minutes- to establish a linear relationship between it and kids getting less sleep and thus performing worse in school.

Jagnani says-

 Using ITUS, I find that an hour delay in sunset time reduces adults’ sleep by 30 minutes. Later sunset also reduces adults’ earnings and household expenditure on education in India
At the time of the survey, Adult earnings increased as one moved West and South- trends which have continued. It is foolish to rely upon a survey which was obviously flawed to arrive at such a bizarre conclusion.
My baseline econometric specification exploits seasonal variation in daily sunset time at the district level after controlling for fixed district-specific characteristics as well as seasonal confounders common across all districts in the sample.

Seasonal variations in sunset time are determined by latitude. There is no need to control for 'district-specific characteristics' because the local administration, though empowered to change School opening and closing times, has no similar power over the Sun or the Earth's rotation.

Jagnani is not making this claim but rather something more bizarre viz that the amount of time a kid has (which is 24 hours) equals some function of the time of sunset plus something to do with the week and the district and an error term.

This is crazy. Electricity access makes a big difference to bed-time for both adults and children. In the dark, there is little to do, except sleep. Very few parts of India ever have a sunset as late as 7 p.m or 7.30 pm with a corresponding 5 a.m sunrise. This still gives 9 hours of darkness for children to sleep. Furthermore, there is no way to differentiate leisure from sleep because kids fall asleep watching TV or listening to a story or reading a book. It is ludicrous to imagine that poor women or children who lacked clocks or watches were able to report the exact amount of time spent playing or sleeping. I myself have a smart watch which claims to tell me how much time I spent sleeping. But it can't differentiate between time I spend reading in bed from the time I am actually asleep.
Sophisticated laboratory based sleep research has revealed that sleep is not itself homogeneous. We need different types of sleep to work effectively. There are non-sleep activities which are a good substitute for the real thing. Equally, excessive sleep of a particular type can have harmful effects.

What Jagnani is doing is Junk Econometrics because it makes a false prediction about the future- viz. that if kids in India sleep less because they go to bed later and are able to spend more time using computers to do their homework, then their academic performance will decline. The opposite is the case.

When I was a kid I was told of legendary scholarship winners from our community who would tie their 'kudumi' (top knot) to a hook in the ceiling so that if they nodded off over their books then they would experience a painful wrench. The popular Hindvi expression 'jo sovath hai, so kovath hai, jo jagrit hai, so pavrit hai'- who sleeps, loses, who is wakeful, wins' is drummed into children with their earliest lessons.
The notion that good quality sleep or equivalent meditative practices are needful is easily accommodated by traditional Indian pedagogy and culture. Indeed, the thing is a matter of common sense.



Using the 2015 India DHS, Jagnani

  compares the outcomes of children of the same age residing in sampling units with different annual average sunset times within a district.
Which Indian district has a one hour difference in sunset time? I suppose Kutch must be the biggest district. It has little more than a ten minute difference in sunset time. Most Indian districts have maybe a 5 minute difference from their Easternmost point to their Westernmost extremity.

Surely it is crazy to write an Econ paper about such a minute difference in sunset time?
Why has Jagnani does so? The answer is that he is assuming a linear relationship whereas common sense tells us there is no relationship at all. If you find an effect, it is an artifact. If you magnify that artifact by ten you get a significant result- but it is impossible to magnify the effect! No district in India has a one hour difference in Sunset time!
Jagnani writes of
exploiting small variation (approximately 5 minutes) in annual average sunset time across the east-west gradient within a district.
His supervisor should have advised him that this was illicit. As is the following assumption-
The underlying assumption when extrapolating the identifying variation to calculate the effect of an hour delay in sunset time is that the relationship between annual average sunset time and academic outcomes is linear.
Wow! What a fantastic discovery! If I want my kid to excel, I should school him in Norway for six months of the year and send him to Chile for their Winter!
I find support for such an assumption. First, as mentioned above, I find that the effect of sunset time on sleep is roughly linear.
If this were true, then kids who shuttled between Norway and Chile would sleep much more on average.
Second, results from the longest laboratory-controlled study on the relationship between sleep and cognitive performance indicate that cognition declines linearly with sleep deprivation
Right! kids who sleep all the time grow up to be Einsteins! Why did my teachers all think I was such a dunce? Probably jealous coz they were Telugu speaking Iyengars in disguise.

Consider Jagnani's findings concerning Andhra Pradesh-
..I use an individual level longitudinal panel from Andhra Pradesh, India, the 2002- 2013 Young Lives Survey (YLS), and show that later sunset is also negatively associated with children’s test scores.
There is only a ten minute difference in Sunset time across the State. Across any given district, it may be 2 or 3 minutes. How the hell can it have a negative association with anything?
This is either monumental stupidity or unconscious fraud.
YLS only provides district-level identifiers and was conducted in a single state in India: Andhra Pradesh. Given these features of the YLS data set, I exploit cross-sectional variation in annual average sunset time at the district level and estimate the effects of annual average sunset on math test scores. I compare the test scores of children of the same age residing in districts with different annual average sunset times within the state of Andhra Pradesh. However, as a robustness check I control for potential omitted variables using lagged test scores as a proxy for child-specific unobservables (e.g., ability) that plausibly co-vary with both annual average sunset time and educational output. I find that children residing in districts with later annual average sunset times have lower test scores. A one standard deviation (25 minutes) delay in annual average sunset time is associated with a 0.3 standard deviation decrease in math scores (Table C.9).
This is a very useful result if 'average sunset time' fluctuated exogenously or if we were speaking of School districts so big that sunset times varied by 25 minutes.
Moreover, controlling for lagged test scores explains meaningful variation in test scores, but does not significantly affect the sunset-test score relationship.44 My point estimate is larger than the effects of later school start times on test scores. For example, Carrell, Maghakian and West (2010) find that a start time of 7 am reduces test scores for a course taught in the first period class by 0.15 standard deviation while a 7.50 am start time has no effect on test scores. Furthermore, with a 7 am start time students performed significantly worse (-0.10 standard deviation) even in subsequent classes, suggesting poor performance throughout the day.
It makes sense to say that 7 a.m feels different from 7.50 a.m. You are less groggy and more alert. But for kids in a class-room setting, there is no difference between 9.30 and 9.35.

Similarly, it doesn't greatly matter whether you are in bed by 9 p.m or 9.05 p.m Yet Jagnani claims-

I find that later sunset causes school-age children to begin sleep later, but does not affect wake-up times.
This is crazy. Nobody knows whether kiddy fell asleep at 9.12 or 9.17- not even the kid.
Even a half hour difference- like that between most of India and the meridian- scarcely matters. This is because there is some variance in the amount of time we spend in bed before falling asleep. That time too is restful.

Countries like the US or Canada do contain States which are so big that Time Zones matter. But this isn't the case in India. Jagnani is exploiting American ignorance to pretend that he has found data from India which supports something possibly of use to a more Northerly country which contains vast, lightly populated, territories.

Jagnani writes-
Emerging out of the British Empire in the mid-20th century, India reckoned a single time zone would serve as a unifying force, and adopted the Indian Standard Time or IST (UTC+5.5) across her territorial boundaries. However, India measures 3,000 km from east to west, spanning roughly 30◦ longitude, corresponding with a two-hour difference in mean solar time.
Hindu India, however, is much more compact. The North East distorts the picture. Indeed, there is a good Religious reason why our meridian runs through the sacred confluence of the Yamuna and Ganga at Prayag.
That corresponds roughly to New York and Utah sharing one time zone, but with a billion more people, of whom 200 million live below the USD 1.90-a-day poverty line (The World Bank, 2016). In such a context, individuals in western India may begin sleep later than those in eastern India due to the relationship between the timing of sunset and bedtime, but fail to compensate by waking up later as India uniformly follows a standard time zone, resulting in potentially large implications for human capital production across the east-west gradient, ceteris paribus.

This is quite mad. 94% of the working population is in the informal sector which is wholly unaffected by rules about clocks made up by bureaucrats. Government run schools have time schedules set by the State Government, with some delegated discretionary flexibility at the local level, but these are all well within daylight hours. Since there is little variance in the number of daylight hours, Time Zones or Daylight Saving is irrelevant for India. We are too far South.


Why pretend that India is highly organised and industrialized like Japan, or that India is a tightly regulated totalitarian country, like Nazi Germany, and that the decision to have just one time-zone hurt millions of little kiddies by depriving them of much needed sleep?
This result is consistent with a model where sleep is productivity-enhancing and increases the marginal returns of study effort for students and work effort for child laborers.
This is a Structural Causal Model. Assuming neither parents nor kids are stupid, the sleep-patterns of children will reflect their preferences. Clocks don't matter because it would be stupid to say 'Sonny boy must reduce his marginal return to study by staying awake till the Clock, set to Indian Standard Time, says its okay for him to nod off.'

Apparently, in India, people in Gujarat are very stupid and don't let their kids have enough sleep because they are slaves to Indian Standard Time. That is why Gujerati has been sinking lower and lower while Arunachal surges ahead.

Jagnani thinks India would be better off if it had two time zones-
I also calculate human capital costs associated with the existing time zone policy against the counterfactual of a two time zone policy: UTC+5 for western India and UTC+6 for eastern India, where western (eastern) India includes districts to the left (right) of 82.5◦ E, and the meridian passing through 75◦ E (90◦ E) defined as its central meridian. In fact, this counterfactual is precisely what was intended by the worldwide standard time zone scheme at conceptions.... India would incur annual human capital gains of over 4.2 billion USD if she switches from the existing time zone policy to the proposed two time zone policy.

India would have zero human capital gains by doing something crazy.  It would lose hundreds of billions. The administration of 5 States would be thrown into chaos. Mirzapur would be an hour ahead of Prayag.  Hundreds of small towns and villages would have a time line running through them. Going about their day, millions of people would have to adjust their watches multiple times.

On the other hand, North Eastern States can and should adjust the times at which Government Offices and Schools start up. But, in law and administrative practice, subsidiarity in this respect has already been mandated for.

Jagnani concludes-
The truth is that most of India is at best half an hour ahead or behind where there 'ought' to be. This is not enough time to generate any significant effect. Whether the Sun sets at Six P.M or Seven P.M does not matter. Everyone gets at least nine or ten hours of darkness. No Indian child gets too little or too much sleep because of Indian Standard Time. Sleep quality is affected the same things which affect quality of Life. However, there may be a case for Public Service Advertisements emphasizing the need for good quality Sleep if some section of the population has irrational beliefs in this regard.
Later sunsets are NOT associated with poorer educational or wage outcomes in India. They are associated with higher wages and educational attainment.

 An article from Live Mint, notes-

'The National Family Health Survey’s (NFHS’) 2015-16 survey of more than 600,000 households shows how clear the East-West divide is. Heatmaps of affluence light up the north-west, west and south-west, while the east and north-east, show poverty and deprivation. 
Bihar, Jharkhand, Uttar Pradesh (UP) and Assam have the highest share of poor households among major states, while a majority of households in Punjab, Haryana, Delhi and Kerala are in the affluent category.
ren’s sleep: when the sun sets later, children go to bed later; by contrast, wake-up times are
not regulated by solar cues. Sleep-deprived students decrease study effort, consistent with
a model where sleep is productivity-enhancing and increases the marginal returns of effort.
Second, sunset-induced sleep deficits have significant negative effects on academic outcomes;
school-age children exposed to later sunsets attain fewer years of education, are less likely to
complete primary and middle school, are less likely to be enrolled in school, and have lower
test scores. Third, later sunsets are also associated with fewer hours of sleep and lower wages
among adults. Fourth, the non-poor adjust their sleep schedules when the sun sets later; the
negative effects of later sunset on sleep are most pronounced among the poor, especially in
periods when households face severe financial constraints.

Clearly Jagnani is an idiot. But, it is Econometrics- as taught at elite American institutions which led him to display his idiocy for all to see. This publication of his is a 'job market paper'.  Clearly, he is considered an up and coming man by the top people in his profession- which is why this paper has attracted so much attention. What it reveals, however, is that Econometrics is a Junk Social Science. It breeds just as great an idiocy as Post Colonial theory or Gender Studies or any other equally incestuous Academic Availability Cascade.

productive effort: later sunset reduces students’ time spent on homework or studying, as well
as child laborers’ time spent on formal and informal work, while increasing time spent on
sedentary and compensatory leisure for all children. This result is consistent with a model
where sleep is productivity-enhancing and increases the marginal returns of study effort for
students and work effort for child laborers.
The second part of the paper examines the consequent
‘lifetime’
or
long-run
impacts of
later sunset on stock indicators of children’s academic outcomes. I use nationally-representative
data from the 2015 India Demographic and Health Survey (DHS) to estimate how children’s
education outcomes co-vary with
annual average sunset time
across eastern and western lo-
cations within a district.
3
I find that an hour (approximately two standard deviations) delay
in annual average sunset time reduces years of education by 0.8 years. School-age children
in geographic locations that experience later sunsets are less likely to complete primary and
middle school, are less likely to be enrolled in school, and have lower test scores.

1 comment:

Anonymous said...

Jagnani is right. Doon School boys are way smarter than Mayo College alumni.