Abstract
Critics of affirmative action policies often claim that increasing the representation of members of marginalized communities in jobs comes at the cost of reduced productive efficiency. This paper reports on a systematic empirical analysis of productivity in the Indian Railways - the world's largest employer subject to affirmative action - in which we examined whether or not higher proportions of affirmative action beneficiaries in employment reduce efficiency in the railway system. We found no evidence for such an effect; and some of our results provide tentative support for the claim that greater labor force diversity boosts productivity.
Affirmative action (hereafter AA) has always been a controversial topic, and the desirability of AA policies is highly contested. This is especially true when it comes to AA in employment. Critics often argue that any possible gains in inclusivity are outweighed by significant costs in economic efficiency. In this paper we report on an earlier study (Deshpande and Weisskopf 2014) in which we subjected this argument to rigorous empirical testing, in the context of a particularly important case of AA that has implications for many similar AA policies around the world.
India has not only the longest history of AA policies but also the most comprehensive system of AA, reaching far more people than all such policies elsewhere. The most prominent form of AA takes the form of “reservations” [i.e., quotas] for Dalits [officially known as Scheduled Castes, or SCs] and Adivasis [officially Scheduled Tribes, or STs]. 22.5 % of all seats in central-government-supported higher educational institutions and 22.5 % of all public sector jobs are reserved for SCs and STs together, corresponding to their share of the overall population in the 1950s – although these quotas are often only partially fulfilled. 1
There are additional quotas for “Other Backward Classes” (OBCs) – i.e., castes and communities low in the socioeconomic hierarchy but above SCs and STs. We focused on the latter because that the AA policy for them has been stable over the time period of our study. See Deshpande (2011) and Weisskopf (2004) for details of India's AA policies, as well as discussion of the debates in India surrounding these policies.
Criticism of AA policies in India is much the same as elsewhere. In the case of AA in employment, it is argued that such policies conflict with considerations of merit because less qualified candidates are selected in place of more qualified candidates, so that poorer quality of work on the job is to be expected from AA beneficiaries. But advocates of AA – in India as elsewhere – argue that hiring is often far from truly meritocratic, and that workforce diversity may actually generate efficiency gains.
To shed empirical light on this debate we focused on the world's largest employer subject to AA – the Indian Railways (hereafter IR), with roughly one and a half million employees. In the US a variety of studies of AA in the labor market have been carried out. A comprehensive survey of those studies – by Holzer and Neumark (2000) concluded that “There is some evidence of lower qualifications for minorities hired under affirmative action programs… ” but that “Evidence of lower performance among these minorities appears much less consistently or convincingly.” In developing countries, however, such studies are very few in number. Most of the studies assessing the impact of AA in India focus either on electoral representation or on higher education. To our knowledge there has not yet been any systematic quantitative study of the effect of AA in the labor market on enterprise efficiency.
The data
For our study we first compiled data separately for 8 regional railway zones, from 1980 through 2002, from various annual reports on productive inputs and outputs. In specifying the variables needed for our analysis, we sought as far as possible to make use of physical rather than money measures, because the latter can easily be distorted by changing prices.
The output produced by the IR consists of passenger service and freight service, measured physically in terms of passenger-kilometers and net ton-kilometers, respectively. We first generated time series indices for total passenger output and total freight output from underlying time series for passenger and freight transport of different types. We then generated time series indices for total railway output by weighting the indices for passenger output and freight output according to their percentage of total railway revenue generated. Since industry outputs are often measured in terms of gross revenue or value added, we also compiled data on railway revenues to obtain an alternative constant -price time series for total railway revenue.
For labor inputs we compiled data distinguishing SCST (i.e., SC and ST) employees from non-SCST employees in four different IR job categories. The upper two job categories are comprised of administrative officers and professional workers; the lower two categories include semi-skilled and clerical staff, as well as relatively unskilled attendants, peons and cleaning staff. To take account of the effects of changes in the job-level-composition of the work force over time, we also constructed time series indices for an alternative measure of total labor input that we call “effective labor.”
For the purposes of our analysis we clearly need to be able to distinguish AA beneficiaries from other employees. The IR provides data on employees who declare themselves to be SC or ST, which is necessary to avail of reservations. Almost all SCST applicants for upper-level jobs do declare their SC or ST status, because they know that their scores on the qualifying exams are unlikely to be high enough for them to gain access to an unreserved job. Any SCST applicant for a reserved job who actually scores higher than the cut-off for a non-reserved job is not included in the count of SCST employees. Thus the data on SCST employment in upper-level jobs measure fairly accurately the corresponding number of AA beneficiaries. On the other hand, many SCST applicants for lower-level jobs do not avail of reservations, because they know they can meet the qualifications for a non-reserved job; so the IR data on SCST employment in lower-level jobs significantly over-estimate the number of AA beneficiaries.
In our analysis we worked with two measures of the SCST percentage: the percentage applying to all employees, and the percentage applying to upper-level employees only. The latter percentage is the better indicator of the effect of AA on IR operations, because it measures the extent to which apparently lower-qualified AA-beneficiary SCST employees have displaced apparently higher-qualified non-SCST would-be employees; and it is this displacement that most worries critics of India's reservation policies.
For capital inputs we chose to work with estimates of gross rather than net capital stock, because measures of net capital stock decline in value as the number of productive future years decline, whereas measures of gross capital stock tend to be proportional to the capital value actually consumed during a given year. We made use of the perpetual inventory method to generate time series of constant-price gross capital stocks. But these measures fail to reflect the extent to which technological progress embodied in capital increases the productive potential of a piece of constant-price capital stock from year to year. So we found it desirable to adjust our raw measure of capital input to take account of changes in capital quality associated with the age structure of capital in order to generate an alternative measure we called “effective capital input”.
The main material input used by a railway system is fuel. Using standard conversion factors to convert all the measures of fuel in physical terms into their equivalent in coal-tonnes, we compiled time series of total coal-tonnes of fuel input. In the case of this input too, we saw reason to generate a second, more nuanced variable to take account of changes in fuel quality associated with changes in the proportions of different kinds of fuel utilized by the IR. The alternative measure of fuel input, which we call “effective fuel,” takes account of the gradual shift toward more efficient diesel- and electricity-powered locomotion.
Production function analysis
We carried out two kinds of econometric analyses to examine the effect of reservations on productivity in the IR. First we did a production function analysis. We started by estimating total factor productivity (or TFP) in each zone-year, using a Cobb-Douglas production-function framework in which output is regressed on labor, capital and material inputs as well as zone-specific unobservable fixed effects and a time variable to capture the effect of technical progress. To check for the robustness of our results, we carried out multiple regressions using our different measures of inputs and output. Moreover, we utilized a variety of different estimation techniques to deal with potential econometric problems – most notably the Levinsohn and Petrin (2003) technique to correct for simultaneity. This refers to the possible correlation between input levels and productivity due to the fact that firms generally respond to anticipated changes in productivity – such as those resulting from shifts in demand – by changing their usage of factor inputs.
In the second stage of our production-function analysis, we assessed the effect of AA on TFP in two different ways. The first was to include a measure of SCST percentage – either for all jobs, or for upper-level jobs only – as an independent variable in the regressions, and then to examine the significance of the SCST percentage coefficient. The second method was to correlate the regression estimates of TFP on a SCST percentage variable, and then to examine the significance of the correlation.
All of our production function results, taken together, reject the hypothesis that higher proportions of SCST employees contribute negatively to productivity levels in the Indian Railways. Indeed, they provide some evidence that higher proportions of SCST employees in upper-level jobs – predominantly beneficiaries of affirmative action – may actually contribute positively to IR productivity.
Data envelopment analysis
As an alternative to traditional production function analysis, we also made use of a quite different method for investigating productivity known as Data Envelopment Analysis (or DEA). This technique requires no a priori assumptions about the functional form of production relations, and it allows for more disaggregation of input and output variables than is possible in production function analysis. 2 Our first step was to use DEA to generate estimates of annual rates of growth of total factor productivity in each IR zone, making use of the effective input variables we had constructed earlier. The second step was to correlate the resulting estimated TFP growth rates with our two SCST percentage variables.
For a thorough explication of the DEA approach, see Ray (2004).
We found that all of the correlations were positive, though in most cases they were not significant at 5 %. There is clearly no support here for the claim that higher proportions of SCST employees result in slower growth in TFP. Where we did find a few positive and significant correlations was with the SCST percentage variable for upper-level jobs. This provides some suggestive evidence in support of the claim that higher proportions of SCST employees in upper-level jobs contribute to more rapid TFP growth in the Indian Railways.
Discussion of the findings
The key findings of our study of the Indian Railways may be summarized as follows: The production function and data-envelopment analyses provide no evidence in support of the claim that higher proportions of jobs filled by SCSTs are associated with lower levels or lower rates of growth of total factor productivity. Furthermore, under some specifications, higher proportions of SCST employees in upper-level positions – who are most likely to be AA beneficiaries – are positively associated with higher levels or faster growth rates of total factor productivity. These findings resonate very strongly with studies assessing the impact of workforce diversity on enterprise productivity in the US, which have found either a positive or null effect, but no evidence of a negative effect.
Our interpretation of the results of this empirical analysis might be contested on the grounds that we have not actually identified the causal relationship at issue. In response to earlier presentations of our work, several people raised criticism along these lines. If the SCST percentage were itself influenced by productivity, or if both these variables were influenced by some other variables omitted from the analysis, then our statistical results could not be interpreted as suggesting the presence or absence of an impact of SCST percentage on productivity. We therefore examined in some detail several specific ways in which the percentage of SCSTs in the IR might conceivably be a function of IR productivity or of omitted variables such as SCST education. It turns out that the institutional processes by which IR jobs are filled – in particular, the limited scope for unit-level managerial initiative in hiring employees – render implausible the causal connections advanced by critics of our work. Thus our examination of these issues ended up giving us greater confidence that we can interpret the statistical findings of our study as shedding light on the effect of India's AA policies on productivity in the IR. 3
For more details on how we have responded to the concerns of critics, see Deshpande and Weisskopf (2014), section 5.
We also addressed concerns that our quantitative measures of IR output – and hence productivity – do not encompass potentially qualitative aspects of IR performance that might be especially sensitive to the competence of railway employees. For example, we undertook a separate analysis of whether higher proportions of SCST labor can be implicated in higher frequencies of railway accidents, as some critics of reservation policies have suggested. Correlating the all-India yearly railway accident rate over the period of our study with the corresponding all-India figures for the percentage of SCST employees in IR total employment, we found statistically significant negative correlations in the case of all employees as well as upper-level employees. The second correlation, which was higher, is the most relevant, since IR employees serving in management and professional positions are especially responsible for guarding against accidents.
Conclusion
In our analysis of an extensive data set on the operations of the world's largest employer subject to affirmative action – the Indian Railways – we found no evidence whatsoever to support the claim of critics of AA that increasing the proportion of AA beneficiaries adversely affects productivity or productivity growth. On the contrary, some of the results of our analysis suggest that the proportion of SCST employees in upper-level positions is positively associated with IR productivity and productivity growth.
Our finding of such positive associations in the case of SCST employees in upper-level jobs is especially relevant to debates about the effects of AA, for two reasons. First, the efficacy with which high-level managerial and decision-making jobs are carried out is likely to have a considerably bigger impact on overall productivity than the efficacy with which lower-level jobs are fulfilled; so critics of AA are likely to be much more concerned about the potentially adverse effects of favoring SCST candidates for upper-level than for lower-level jobs. Second, it is precisely in the upper-level jobs that reservations have been indispensable for raising the proportion of SCST employees.
The results that we have obtained from our analysis of productivity in the Indian Railways are consistent with results obtained from productivity studies in the US, in that there is no statistically significant evidence that AA in the labor market has an adverse effect on productivity. Our results are stronger, however, in that we do find some suggestive evidence that AA at the upper levels of the labor market actually has a favorable effect in contributing to greater productivity.
It is beyond the scope of this paper to explain just how and why AA in the labor market might have such a favorable effect. We can, however, adduce some relevant evidence from research carried out by others. A number of studies in India and elsewhere indicate that hiring practices are often far from meritocratic in the absence of AA. For example, in a study of modern urban Indian highly-skilled labor markets for private sector jobs (often assumed to be among the most meritocratic), Deshpande and Newman (2007) have shown how caste and religious affiliations of job applicants shape employers' beliefs about their intrinsic merit. This confirms the findings of other studies that uncover labor market discrimination and point out how social identities impact hiring and wage offers.
Furthermore, there are numerous a priori reasons to expect that AA in hiring might improve economic performance – particularly in high-level jobs. Individuals from marginalized groups may well be especially highly motivated to perform well when they attain decision-making and managerial positions, because of the fact that they have reached these positions in the face of claims that they are not sufficiently capable; they may therefore have a strong desire to prove their detractors wrong. Such individuals may simply believe that they have to work doubly hard to prove that they are just as good as their peers, so they may actually work harder. Furthermore, to have greater numbers of managers and professionals from disadvantaged groups working in high-level positions might well increase productivity because their backgrounds make them more effective in supervising and motivating other workers from their own communities. 4 Finally, improvements in organizational productivity may well result from the greater diversity of talents and perspectives made possible by the integration of more members of marginalized groups into high-level decision-making teams. 5
This recalls the arguments in favor of AA in U.S. educational institutions made to the Supreme Court by U.S. military officers, who want to avoid having just white men in charge of troops that are disproportionately of color
Page (2007) has shown convincingly how groups that display a wide range of perspectives outperform groups of like-minded experts.