Big Brother is Watching

This paper is not about the NSA, but the more mundane surveillance that takes place in workplaces every day. Drs. Lamar Pierce, Daniel Snow and Andrew McAfee have published a new study of the effects of surveillance of employees in the restaurant business using informational technology (IT).

The setting is so-called “casual dining” restaurants. The authors don't identify the five chains they studied, but say they come from the market that includes Applebee's, Chili's and Olive Garden. Employee theft in the restaurant industry is estimated at 1% of revenue. The management of these chains installed Restaurant Guard, a computer software product sold by NCR designed to identify suspicious transactions by servers (waiters and bartenders). The sample was large: 392 restaurants in 38 states, employing over 30,000 servers, who conducted about 630,000 transactions per week.

The authors don't explain how the IT product works, which is probably a trade secret, but give examples of suspicious transactions. A voided check is regarded as suspect because one way servers steal is to void the check after the customer has paid and pocket the money.

The study uses an interrupted time series design. The main problem with before-after designs is the possibility that some unknown outside events coincide with the treatment and influence the results. In this study, the IT was installed on a staggered basis over a two year period (March 2010 to February 2012), protecting the results from what the authors call “week-specific shocks” (but not from long-term economic trends).

The researchers had no control over how the restaurants used the information about suspicious transactions. It is their understanding that they usually informed servers when Restaurant Guard was installed in order to take advantage of its deterrent effect. The restaurants agreed to apply the software retroactively to the time before it was installed, which allowed the researchers to measure changes coinciding with the intervention. Here are the main results:

• Theft losses averaged only $108 per restaurant per week before the intervention, and were reduced by 22%, or $23 per week. Either there was not much theft or the IT was not detecting all of it. (The authors assume the latter.)
• The impact of IT monitoring on total revenue was substantial. Revenue from food sales increased 7%, or $2975 per restaurant per week. Drink sales increased 10.5%, or $927 per restaurant per week.
• Tip percentage could only be measured on credit card transactions, so we must assume that the results were similar when customers paid cash. Tip percentage was basically unchanged. It increased by .3%, going from 14.8% of the bill before the intervention to 15.1% after.
• The researchers analyzed the results for individual workers. The effect of the surveillance was fairly uniform across workers. There were no significant differences in behavior between “known thieves”—servers whose behavior was flagged as suspicious—and “unknowns” who had not behaved suspiciously. There was more attrition among the “known thieves,” but it was seldom traceable to specific incidents, suggesting that they were leaving voluntarily rather than being fired.

If Restaurant Guard was not identifying much theft and few workers were fired, why did revenue go up sharply? The authors speculate that before the IT intervention, workers were “multi-tasking;” that is, both working and stealing. The software had the psychological effect of increasing fear of detection and discouraging theft. The workers compensated by increasing their efforts to make sales—for example, by asking customers whether they wanted another drink—in order to compensate for lost theft income by increasing their tip income.

The circumstances suggest the possibility of a Hawthorne effect, or reactivity, in which participants change their behavior due to the awareness that they are being observed. This effect might disappear over time as they gradually forget about the intervention. The authors compared changes in behavior during the first three months following the intervention and found no systematic trends. I'm not sure that three months is long enough to measure the decline of a Hawthorne effect.

The other main conclusion the authors draw from the study follows from the relative absence of individual differences among workers, whether “known thieves” or “unknowns.” The conventional wisdom is that you protect yourself from theft by hiring honest workers and firing those who turn out to be dishonest. But they conclude that employee theft is influenced more by environmental factors—in this case, surveillance—than by worker traits.

This study, and the role of social scientists in conducting it, makes me uncomfortable. Restaurant workers are some of the most dramatically underpaid workers in our society. The minimum wage for servers is $2.13 per hour. The Restaurant Opportunity Center, which advocates for better working conditions in restaurants, reports that the average restaurant worker earns $8.89 per hour (including tips), for a total of $15,092 per year; 89.7% don't have health insurance; and 87.7% don't have paid sick days. While theft is an inappropriate response to this situation, it is not surprising that they steal and that some of them feel justified in doing so.

The authors, all business school professors, adopt a somewhat self-congratulatory tone at having induced these servers to become more productive—to stop stealing (if they were actually stealing) and work harder for the same low wages. The increased value of their labor went almost entirely to the restaurant owners. Tip percentage increased by only .3%. The main way these servers profited was from the fact that their tips were based on larger checks.

The authors did not attempt to measure this increased tip income; in fact, there's no evidence that they even cared about it. It's impossible to determine tip income per server accurately from the data they provide. I did some rough back-of-the-envelope calculations making what I thought were reasonable assumptions and estimated that the average server in the study, who worked 22 hours per week, netted about $15 a week in additional tip income. I would hesitate to call this a “raise,” since it was obtained by working harder.

This study is consistent with theory and research in social science that suggests that one of the effects of new technology is to increase inequality. New technologies, such as computers, increase the income of the wealthy, who are able to take advantage of them, relative to those who are unable to afford the technologies or who are not trained to use them.

It's unfortunate that these social scientists are not interested in looking at the causes of illegal behavior among the wealthiest 1% of Americans. I guess there's no corporate financial support for that kind of research.

You may also be interested in reading:

Raising the Floor

Catch-22