Chris Stanton, a PhD candidate at Stanford Business School, is using oDesk data in his research. Below, he shares some basic economic insights about the oDesk market.
In this post, I will concentrate on the role of feedback on provider wages. I hope these results, coupled with previous posts on the returns to tenure and training, help providers form expectations about long-run earnings trajectories. Overall, the results suggest that providers who receive good feedback and gain experience on oDesk can receive significantly higher wages over time. I find that a change in feedback score from 2.5, the mean score in the data, to the maximum score of 5, results in wages that are about 5.4% higher.
While oDesk users surely expect a positive relationship between feedback and provider quality, quantifying the economic effect of feedback on wages is statistically tricky. The difficulty arises because the best providers are likely to get the best feedback, but these same top-notch providers are also likely to have unobserved attributes like superior interviewing skills that simultaneously result in high wages. I use a statistical procedure to account for unobserved provider skills.
The data covers matched assignments on oDesk from the platform launch until May 2008. This includes observations on 7,123 providers matched to 28,321 assignments. The description of my statistical strategy may be esoteric, so the casual reader may wish to skip to the results section. The basic idea is that I use fixed effects multivariate regressions to control for any time-invariant provider characteristics which may be correlated with a provider’s feedback. Because I am able to identify how changes within a single provider’s feedback influence his or her wages over time, this strategy addresses unobserved provider characteristics which otherwise hamper the measurement of the effect of feedback on wages. In my preferred specification, I regress the logarithm of hourly wages on a polynomial of the provider’s weighted feedback score, time using the platform, and overall platform time trend. I also include controls for the number of tests a provider has taken.
In this first graph, the overall effect of feedback on wage percentage changes is given in blue. Separate results for Indian and Russian providers are also provided. Not surprisingly, the results show that relative to having zero feedback, providers with low scores do slightly worse. On the positive side, providers can expect 2.5%-5% more earnings from the best feedback scores.
The second graph breaks out the effect by job type. The graph is a bit difficult to interpret because the effect of feedback on wages for writing jobs appears huge. But be warned – the effect for writers looks large in the sample, but is not statistically significant. Web and software developers can expect to earn about 5.6% more with a feedback score of 5 versus 2.5, while providers of administrative support earn even larger percentage increases with good feedback.