Sunday 6 August 2017

Corrupt Indians prefer Government jobs, Stupid Academics prefer Junk Social Science

Hanna & Wang, of Harvard & Wharton respectively, have a paper claiming that propensity to corruption is linked to a preference for Public Sector employment in Karnataka, India.
Students in India who cheat on a simple laboratory task are more likely to prefer public sector jobs. This paper shows that cheating on this task predicts corrupt behavior by civil servants, implying that it is a meaningful predictor of future corruption. Students who demonstrate pro-social preferences are less likely to prefer government jobs, while outcomes on an explicit game and attitudinal measures of corruption do not systematically predict job preferences. A screening process that chooses high ability applicants would not alter the average propensity for corruption. The findings imply that differential selection into government may partially contribute to corruption. 
Hanna & Wang didn't survey Arts students- claiming people who study Politics or History or Literature don't end up as Public Sector workers even though, historically, Arts subjects are considered a gateway to Babudom. Instead, their sample is about 80 percent Commerce and 20 percent Sciences. Students of the former receive some instruction in basic concepts of Game Theory and Mechanism Design and this is reinforced by popular articles. It is likely that such students would show different behaviour in a 'lab test' even if they share the same normative values with students who don't understand the importance of the concept of Mechanism Design in modern Social Science. In other words, this methodology is ab ovo flawed.


Do Hanna & Wang have 'a screening process' for high ability applicants? No. Their sample was of students, not applicants for UPSC jobs. If the cost of application for Public Sector jobs is zero, which is what they assume, then, okay, there is likely to be a correlation. However, the actual cost of attempting the UPSC exam in a credible fashion is at least 700 dollars in coaching fees and other expenses plus an enormous opportunity cost in terms of time and effort. It is unlikely that the 80 percent Commerce stream, 60 per cent female, 40 per cent minorities, sample they have used is in any way representative of actual applicants, let alone successful ones.

This is not to say that a survey of actual applicants for UPSC jobs might not yield a similar result. However, our two researchers have not in fact made any such survey. Yet, by a piece of verbal sleight of hand, they substitute 'applicant' for 'student' on the basis of a farcically wrong assumption. Their work does not 'predict' anything- save perhaps that their next paper will be garbage.

Hanna & Wang's methodology was as follows-
 'we asked each student in our sample to roll a standard die 42 times and to report the number of each roll in order to receive a payment that was increasing in the number reported. Thus, while we do not know with certainty if an individual lied, we can observe how far each individual’s distribution of reports is from the uniform distribution. Note that this measure is appealing in that it does not prime the subject on corruption or dishonesty explicitly and allows them to feel comfortable in knowing that no one can say with certainty that they are cheating.                   

'...one key contribution of our paper is that we then conducted a validation exercise of this measure using a real measure of corruption. Specifically, we conducted the dice task with 165 government nurses who were part of an experiment conducted by Dhaliwal and Hanna (2013), in which they collected detailed measures of absenteeism through the use of random checks over two years. Thus, we can test whether the dice task outcome predicts fraudulent absenteeism. '

So students aiming to crack the IAS are being compared to poorly educated female nurses in the districts who have much lower cognitive ability and who are being paid for their participation in the 'lab test' with candy, not money.

Why? It is because the main job of an IAS officer ( 75 % of whom are likely to be male) is emptying bed pans. Moreover, a young Bureaucrat who is good at his job can easily get a visa to Europe or the Gulf and earn much more money for doing the same work.

What about 'fraudulent absenteeism'? Is that a good measure of propensity for corruption? No. A corrupt staff nurse will show up for work every day and extort bribes from patients by threatening to withdraw vital nursing services. Moreover, absenteeism is high where the posting is undesirable- some rural shithole- but the posting is undesirable only because the nurse doesn't have the money or the connections necessary to pay a bribe for a good posting. Such absenteeism has to be tolerated because of poor working conditions and the fallibility and expense of monitoring.

There is literally no similarity or connection between the two sample populations Hanna & Wang have chosen. The nurses are poor and stupid and what's more they know that aint going to change any time soon. Their fate is sealed. The students are not as poor or as stupid and their life is still before them. They can dream dreams. With hindsight, it will be obvious that a lot of them hadn't a hope of cracking the UPSC exam. Some may end up with MPhils or PhDs but still applying for peon's jobs.

Why are Hanna & Wang pretending that there is some hard-wired trait corresponding to a stable, exogenously given, propensity for corruption? Is there any evidence that the thing exists? If it does, why not study how it can be detected and cured? But, why stop there? Why not subscribe to a wholesale Manichaeism? Why not say that this survey predicts that the evil Demiurge is recruiting Satanically inclined minions to staff the Public Sector in a manner that passes Human Understanding?

For the moment, Hanna & Wang are being modest. Probably, they suspect they aren't 'high ability' and so will be screened out if they start babbling about Gnostic Demiurges or Lizard People from the Planet X. Thus they observe-
 students who had above median dice points (i.e. higher probability of cheating) were 6.2 percent more likely to want a government job. We find no significant difference in the predictive value of the dice task for high-ability students than for low ability students in terms of job preferences. This implies that screening on ability would neither exacerbate nor mitigate the selection problem among government workers in this context. Importantly, we find that nurses with above median dice points were 10.7 percent more likely to be fraudulently absent than those below it. Furthermore, as in the student sample, we do not find any significant heterogeneity in the predictive power of the dice task for nurse absenteeism by ability. This simple fact that the dice task also predicts the corrupt behaviour of the government nurses helps validate the interpretation of the student sample. 
Liars tell more lies than non liars which is why it is meaningful to speak of liars as opposed to non liars. We expect Public Sector workers who lie about one thing to be more likely to lie about another thing. Does this prove that liars are more likely to want to be Public Sector workers? No.

Perhaps it 'validates an interpretation' of a study which shows liars prefer Public Sector jobs? No. Substitute the phrase 'Cat lover' for 'Public Sector'. The fact that Cat loving liars tell more lies than Cat loving non liars does not 'validate an interpretation' of a small sample test where it was found that Cat lovers were 6.2 % more likely to tell lies. On the contrary, it invalidates any interpretation of the academic paper in question as other than Junk Social Science.

Civil servants need to be civil. They slot into a hierarchy. It may well be that telling lies is correlated with a type of 'Social intelligence' which is based on what the Greek Church calls oikonomia as opposed to Akribeia- i.e. flexibility rather than rigid exactitude. Lord Armstrong, a leading Thatcher era British Civil Servant, is credited with popularising the expression ' to be economical with the truth'. There is no suggestion that Armstrong was corrupt. Why not? Well, Armstrong was subject to strong checks and balances. These could only be subverted by a powerful nexus created and sustained by the elected politicians he answered to. Even then, he'd have been caught unless the British Courts had as big a backlog of cases, or were as accommodating in granting extensions, as Indian Courts.

In India- Karnataka specifically- preference for Govt. employment is tied to housing situation and family obligations. Those who are well housed in a good area are also likely to have better educational opportunities and choose high paid private sector employment. The Public Sector is only sought by either dunces or those who are part of a powerful clan- i.e. they are, in effect, being recruited by an existing syndicate. One reason for this is that the first years of Government service are financially strenuous. One needs to pay a lot in bribes to get a good posting, nice housing, a non crazy rapist of a boss etc. At this time, it is the extended family which supports you. Sometimes, it is the in-laws who stipulate that you stay in such and such Govt. job so as to be useful to them when you get promoted on the basis of seniority (the bribes only affect posting, not status).

Hanna & Wang chose to remain blissfully oblivious to any and all considerations of the sort listed here. Why? The answer is that Stupid Academics like telling Stupid Lies. Other Stupid Academics, like the good folk at Marginal Revolution, are happy to quote these Stupid Lies because that's how availability cascade based Junk Social Science operates.

No comments: