Incentives and service quality in the restaurant industry: the tipping–service puzzle

Abstract

Tipping is a significant economic activity (tips in the US food industry alone amount to about $42 billion annually) that was claimed to improve service quality and increase economic efficiency, because it gives incentives to provide excellent service, and therefore allows to avoid costly monitoring of workers. The article suggests that this common wisdom might be wrong. A simple model shows formally that tips can improve service only if they are sensitive enough to service quality. Empirical evidence suggests that tips are hardly affected by service quality. Nevertheless, rankings of service quality by customers are very high; the co-existence of these two findings is denoted ‘the tipping–service puzzle,’ and several possible explanations for it are offered.

Acknowledgement

I am grateful to an anonymous referee for helpful comments.

Notes

¹ As mentioned in the previous section, some of the literature on service quality suggests that perceived quality by the customer depends on the gap between expected and actual service quality (e.g. Parasuraman et al., 1985, Li et al., 2003). Here I use q for service quality, without incorporating in the model additional variables that relate to the question how exactly q is formed. Consequently, there are two possible interpretations of q, and the reader can choose the one he prefers. First, q can be interpreted as perceived service quality that is equal to the gap (actual service quality–expected service quality). The waiter being able to choose q can then be the result of assuming either that different customers of the same restaurant have similar expectations about service quality, or that customers are heterogeneous in their expectations but the waiter is able to identify the service expectations of the customers. Both alternative assumptions lead to the result that the waiter knows the one-to-one correspondence between the actual service he provides and q, which implies that he can choose q. A second interpretation of the model is that only actual service quality affects tipping (expected service quality does not) and q is interpreted as actual service quality. Notice that this still allows the customer to have expectations about service quality and to be disappointed or surprised with the service he receives. For example, a customer who expects excellent service might plan to tip 20% of the bill. Another customer expects only mediocre service and therefore plans to tip 15%. Eventually both receive good service. The first customer is disappointed, so he tips less than he planned, say 18%, and the second customer is positively surprised and therefore he tips more than he planned–also 18%. This shows that tipping may be a function of only actual service quality even if customers also have expectations about service quality and their utility is affected by these expectations (through disappointment or positive surprise). There are several good reasons to adopt q as service quality without further complicating the model by introducing two variables, one for expected service and one for actual service. One reason is that it allows to keep the model simpler, the analysis more tractable, and the results more intuitive, simple and convincing. A second reason is that the empirical evidence that is analysed in the following section based on the model has only one measure of service quality–the score that the customer gave to service quality after the dining experience. The data in the various empirical studies that are analysed (and in any other tipping study I know of) does not have a separate measure of expected service quality, so introducing into the model a variable that has no equivalent in the empirical data it aims to analyse seems pointless. Third, this modelling is in line with other theoretical models of tipping, where service quality is measured as a single variable and not as a gap between expected and actual service (see for example Azar, 2004a, 2007c).

² The reason that people want to feel generous and kind is that it can improve their self-image, which might affect their utility, see for example Akerlof and Kranton (2000) and Loewenstein (1999).

³ In that case the customer should not tip, but empirical evidence shows that almost everyone tips when this is the norm (Azar, in press), and therefore it is reasonable to assume that the representative customer has at least one reason to tip.

⁴ The second-order condition is given by t″(q*) − e″(q*) < 0. In what follows I assume that this condition is satisfied. It is easy to see that this is the case if we assume that t is concave or linear. The second-order condition may be satisfied, however, even if t is convex, as long as e is ‘more convex.’

⁵ Taken from the full sample results with service quantity included in the regression, Table 3 in their article.

⁶ This information is based on the dataset itself and not the articles. I thank Orn Bodvarsson and William Gibson for their data and Mike Lynn for help obtaining the data.

⁷ The authors discard certain observations such as tables with more than 5 patrons (because in this case the tip is often included in the bill already) or tables with multiple checks.

⁸ Notice that the negative coefficient of Knowledge of menu suggests that if the aggregate measure of service quality included also Knowledge of menu in the average computed, the coefficient of the aggregate measure would have been smaller.

⁹ The federal law requires employers to pay $5.15 per hour in general. Tipped workers should also have total income (from wages and tips) of at least $5.15 per hour, but their wages can be reduced up to $2.13 an hour, using what is called ‘tip credit’ towards the $5.15 minimum wage. Some states adopted different laws, however.

¹⁰ One exception is the study of Bodvarsson et al. (2003), who argue that using different econometric techniques (instead of OLS regression) in their sample of 247 diners in a Central Minnesota restaurant reveals a stronger relationship between service quality and tips than in other studies. Lynn (2004), however, re-analyses their data and suggests that the effect of service quality is small (top-rated service increases tips by not more than 1.5% of the bill amount) and not statistically significant, in line with previous research (see also Bodvarsson, 2005 for a reply to Lynn's criticism).

¹¹ On psychological motivations that play a role in job satisfaction see Hackman and Oldham (1976) and the other literature about Job Characteristics Theory; on job satisfaction in the low wage service sector, including the restaurant industry, see Brown and Mcintosh (2003).