a. Non experimental evidence
Our work on the WMS data fits within a large body of literature examining the effects of management on firm performance. Several findings support the hypothesis of a positive relationship between management quality and firm performance. First, correlating the same summary management-quality measure underlying Figure 3 with various firm-performance outcomes suggests that higher management scores are positively and significantly associated with higher productivity, firm size, profitability, sales growth, market value, and survival in the manufacturing sector. For example, Figure 5 shows the local linear regression of log of firm sales on the management score. Since we would expect the better-managed firms to capture a larger fraction of sales, the positive and monotonic relationship is consistent with this prediction.The relationship between productivity and management is robust to different ways of combining the management questions, and to controlling for firm-specific, time-invariant characteristics using the panel dimension of the management data. Fixed-effects estimates of the management coefficient are indeed also positive and significant, although the magnitude of the association is smaller.
Fixed Effects (FE) ModelFE is a widely used econometric tech-nique that exploits the time dimension of repeated observations for, say, the same individual, to account for any time invariant and individual specific characteristic. In this way, the result-ing estimate can be interpreted as the causal effect of the variable of inter-est. Note, however, that any time vari-ant occurrence, such as a random shock, that affects the variables in your model, represents a real threat to the interpretation of the results.
The association of management with organizational performance is also clear in other sectors outside manufacturing. Bloom et al. (2010) finds that management scores in a sample of orthopedic and cardiology departments of UK hospitals are significantly associated with better patient outcomes. Chandra et al. (2013) show that there is also a positive association between case-mix-adjusted AMI (heart attack) survival rates and management scores among hospitals in the United States. In subsequent work, Bloom, Sadun and Van Reenen (2013) show that this positive relationship between patient outcomes and management also holds in other countries.
In the six countries for which we have school-level pupil outcome data (the United Kingdom, the United States, Sweden, Brazil, India, and Canada), there is again a positive and monotonic relationship between pupil test scores and the management scores of the schools, as shown in Figure 6.
In recent years, a number of studies using the WMS methodology have corroborated the finding that management scores are positively associated with measures of organizational performance. One exception, however, is the Rasul and Rogger (2013) study of the Nigerian civil service, which examines the success rates of 4,721 projects, such as plans to build dams and roads. After implementing a survey mirrored in the WMS methodology, they found that, contrary to the other studies, organizations with high management scores were less likely to successfully complete projects. By contrast, decentralization was found to be associated with a greater likelihood of project success. The authors’ preferred explanation is that the greater monitoring associated with higher management scores crowds out the intrinsic motivation of the public servants.
b. RCT evidence
A problem with the non-experimental evidence is that management is likely to be endogenous. Even in the panel estimates, time-varying unobservable factors may be correlated with both management and performance. Reverse causality may also be an issue: perhaps better-performing firms can employ superior management consultants, for example. Hence, in recent years an emphasis has been placed on randomized controlled trials (RCTs) to obtain causal estimates.
Randomized Control Trials (RCTs)As for today, RCTs are considered the gold standard of applied economics and other disciplines. The reason why researchers performing empirical analysis appreciate RCTs has to do with the several problems that they face when attempting to infer a causal relationship between the variables under study. Let’s imagine, for example, that a researcher wants to assess the impact of a training programme offered to unemployed people on the ability to find a job afterwards. If this researcher will estimate the casual impact of the programme by comparing position secured after a month between people who joined the programme and people who did not, he or she will very likely estimate an invalid impact. In fact, people who decided to join the programme has some unobservable characteristics that determine their success in finding a job with respect to people who did not join, such as greater motivation. RCTs overcome this issue (known as selectivity bias) randomising the allocation of the “treatment” (the programme, in our case). If the sample is large enough, people who were randomly assigned with the treatment do not differ, on average, from people who were not allocated with the treatment — the resulting discrepancies in outcomes (finding a job, in our case) are interpreted as the causal effect of the treatment. Note, however, that RCTs may still suffer, among other problems, from small sample size, poor take-up, contaminations between groups and the Hatwhorne effect.
In the manufacturing sector, an RCT run by Bloom et al. (2013) provides important contribution in the study of the causal impact of management on firm performance. In this study, the research team provided free management consulting to textile plants in India to help them adopt the kind of modern management practices measured by the WMS. The researchers compared the performances of two sets of randomly selected plants: those that received the consulting and the control group that did not. The experiment revealed that the adoption of these management practices led to large increases in productivity: a one standard deviation increase in the management score increased productivity by 10 percent. This figure lies between the OLS levels cross-sectional and within-groups panel estimates in Bloom, Sadun, and Van Reenen (2016). Profits in the first year increased on average by $325,000, which compared to a market cost of the intervention of $200,000. So, the intervention more than paid for itself in the first year. The fact that the improvements seem to have persisted suggests that the total returns will likely be even higher.
Interestingly, the Indian experiment also found that the adoption of these types of practices was more likely to occur when firms were struggling. When facing tough times, firms were more likely to try to upgrade their management practices. In contrast, when conditions were better, firms were reluctant to change or adjust management practices. If this type of endogeneity were common, it would lead to systematic underestimation of the impact of management on performance, at least in panel data estimates that rely on changes in performance following changes in management.
A growing number of RCTs have also studied management interventions in developing countries in micro-enterprises (single- or few-person firms). The results of these are much more ambiguous than those from the Indian textile experiment (which, by contrast, focused on large firms). Karlan, Knight, and Udry (2012) survey 11 studies of managerial interventions. Several of these find positive effects on profits, results that are similar to those of the Indian textile RCT. These RCTs include Mano et al. (2011) in sub-Saharan Africa; Valdivia (2012) in Peru; and Bruhn, Karlan, and Schoar (2012) and Calderon, Cunha, and De Giorgi (2013) in Mexico. Others find insignificant or mixed results; Berge et al. (2011), for example, find positive effects for men but negative effects for women. Some other studies find negative effects. Among these are Gine and Mansuri (2011), and Drexler, Fisher, and Schoar’s (2011) basic accounting training. These studies are summarized in McKenzie and Woodruff (2012).
Several possible factors may explain why the wider literature does not find uniformly strong and positive effects such as the RCTs of Bloom et al. (2013). First, the Bloom intervention (like the WMS) emphasizes formal systems for monitoring output, inputs, and defects; setting short- and long-run targets; and establishing rigorous employee appraisal systems. These are less likely to be important for the micro- and mini-enterprises—mostly single-person firms. The Indian textile RCTs (and the WMS survey) explicitly target larger firms with several hundred or thousands of employees spread across multiple factories. Second, the firms that deliver the management consultancy services in the wider literature are usually local firms, unlike Accenture, the global firm that delivered the services for the Indian experiment. Such local firms may struggle to deliver the same quality of intervention of global consultancy firms. Third, the types of management training differ substantially. The WMS method focuses on operational improvements, whereas many of the treatments focus on “strategic management,” such as improved marketing and pricing. Consistent with the latter two points, McKenzie and Woodruff (2016) show that the measured effects of training on profits and sales are consistent with the magnitude of the changes in management practices observed following the training interventions. The problem is that the training programs aimed at smaller enterprises result in only very modest changes in management practices. This suggests the need to focus on both the content and the quality of delivery of the training. And while the RCT closest to the WMS approach (the consulting experiment in India) does find causal effects consistent with the non-experimental work, understanding the heterogeneity of the effects across different RCTs is therefore an important area for future research.
