Here I report briefly on a recent paper in the Cambridge Journal of Economics by Haile, Sadrieh and Verbon. I appreciate the intentions of the paper, but I believe that it was poorly executed and, consequently, their results are unconvincing. As the authors state, "[N]o study so far has attempted to disentangle the ethnicity effects from the income inequality effects." (705) Their study fails to disentangle the effects too.

The authors ran trust games in two South African universities: The Universities of Potchefstroom and Mafikeng. (Clarification: They are now both campuses of North-West University.) They say the University of Potchefstroom is predominantly White, it is not. In 2003 the university had 41% White students, and the majority of students were Black (47%) with the remainder Coloured, Asian and Other (report here). For Mafikeng the statistics in 2003 were 0.5% White and 98.5% Black with the remainder Coloured and Asian (report here).

The authors ran anonymous control experiments in which no information was revealed. In the treatment experiment both race and income level were revealed. They revealed race and income by telling subjects directly whether their partner had reported themselves as White or Black, and also whether their partner had reported their perceived income as above average or below average. There are thus four potential categories: {Black-Low, Black-High, White-Low, White-High}.

The first problem is with their racial categorization. In South Africa people report whether they are White, African, Coloured, or Asian with the three final groups all counting as 'Black' (see the CEE report here). The authors assumed that all subjects would know what was meant by 'Black'. I did not see evidence that they tested this. Moreover, as shown by Tyson et al (1988), the levels of prejudice within Whites varies dramatically with significantly lower levels of prejudice (towards non-whites) among English home language Whites than among Afrikaans home language Whites. Furthermore, when the researchers provided the racial information in this way it erased any possibility to understand the underlying mechanisms to identify ingroup and outgroup membership. Why not use names (Bertrand and Mullainathan 2004, Van der Merwe and Burns 2008)? photographs (Eckel and Wilson 2002, Burns 2004, 2006)? cartoons (can't remember the reference)?

Second, their experimental design was poor. To separate the effects of race and income, and to postulate about combined effects we need experiments on each of these separately. What this means is that they ran an insufficient number of treatments. We not only need race treatments alone, but also income treatments alone for a full complement of treatments that would allow valid interpretations of their results, i.e., we need {(black, white) (white, black), (black, black) (white, white)} treatments, and {(high, low) (high, high), (low, low), (low, high)} treatments. Once these have been run then you can do Haile, Sadrieh and Verbon's experiment. Their experiment does not actually isolate what they say it does.

Third, saying that low income Blacks don't give to high income Whites is because of 'envy' may be inaccurate. Though the behaviour is exhibited for low income blacks towards high income whites only they may feel resentment, spite, or hatred towards this group for reasons correlated with high income and whiteness (say, family complicity with apartheid, perceptions of persistent racism, political inclinations, undeserved benefits - on average Whites more likely to be employed and have higher income in North-West than Blacks). Black people in Potchefstroom, for example, may hate the (predominantly Afrikaans?) richer Whites. I understand that saying hatred motivates people is not politically correct, but saying that the Haile, Sadrieh and Verbon result is a consequence of envy without admitting other possibilities is speculation. What about disgust? spite? anger?

Fourth, why do they not present any independent evidence from the sampled students, or perceptions by other students at the universities, on socio-economic data about black, white, high income and low income individuals? If we had more information about these factors then we might have greater insight into the factors underlying the 'envy' result.

Lastly, the sample does not contain a sufficient number of wealthy Blacks. I believe this to be a consequence of where the authors chose to run the experiments. Had they chosen to run the experiments in Cape Town, Johannesburg or Pretoria they would probably have found more (though not many) wealthier Blacks. Instead, they chose areas that were particularly unsuited to their needs. North-West is one of the poorer South African provinces (see basic provincial information).

There are several other points worthy clarifying, but the points I raise were those that came to mind. I hope that more and similar work is carried out in South Africa, particularly work that "[attempts] to disentangle ethnicity effects from income inequality effects." With this work as a baseline I think that a substantial amount of research can proceed and clarify the results that Haile, Sadrieh and Verbon present.

Point of clarity
I have tried twice to contact Prof Verbon as the contact author listed on the paper to obtain additional information not listed in the paper. I have not received any responses.

Reference
D. Haile, A. Sadrieh, H. A. A. Verbon (2008). Cross-racial envy and underinvestment in South African partnerships Cambridge Journal of Economics, 32 (5), 703-724 DOI: 10.1093/cje/ben011

Additional Sources:
SouthAfrica.info The all-in-one official guide and web portal to South Africa.
A StatsSa (2002) report on North West from 2000 data.