The impact of customer satisfaction on CEO bonuses

In an unpublished study, Srinivasan et al. (2003) find that satisfaction—as measured by the American Customer Satisfaction Index—has no incremental explanatory power for CEO compensation in their study of the airline industry. However, their work focuses on a very small sample drawn from a single industry. In another unpublished paper, Chen et al. (2008b) predict and find statistically significant regression coefficients for terms representing the interaction between satisfaction and various proxies for industry competition, but report no significant “main effect” association between CEO bonus and satisfaction. Indeed Davila and Venkatachalam’s (2004) study, which demonstrates that passenger load factor is a positive and significant determinant of CEO bonus in the airline industry, is the only evidence showing a direct statistical association between any nonfinancial measure and CEO compensation in any sector.

Customer satisfaction impacts each of the financial metrics (Anderson et al. 2004; Gruca and Rego 2005). Therefore, testing the incremental independent impact of U(ACSI) provides a conservative estimate of the relationship between customer satisfaction and compensation. Similarly, studies by Jacobson and colleagues investigating the incremental information content of marketing metrics with respect to firm value (e.g., Aaker and Jacobson 1994; Mizik and Jacobson 2008) provide a conservative estimate of value relevance.

Separately, Dikolli and Sedatole (2007) and Chen (2009) demonstrate that the benchmark against which satisfaction is measured influences its information content.

As discussed, consistent with standard practice for studies that employ an expected/unexpected framework, U(RET) = RET throughout.

We are grateful to the anonymous review team for their suggestions with respect to this test.

The mean for U(ACSI) for the CLEAR SIGNAL observations is .677 (N = 536) while that for the AMBIGUOUS SIGNAL observations is -1.04 (n = 212) and the overall mean is .18 (N = 748). A t-test of the difference between the means for the CLEAR SIGNAL and AMBIGUOUS SIGNAL subsamples is significant at p = .000 with a t-statistic = 6.25. This finding is confirmed by Hotelling and Kruskal-Wallis tests. These findings suggest that U(ACSI) is significantly different for the AMBIGUOUS SIGNAL observations. No other significant differences in the mean values of variables across the CLEAR SIGNAL and AMBIGUOUS SIGNAL subsamples are evident.

For each individual regression estimate outliers defined as those observations with values of Cook’s Distance greater than 4/N are excluded from the analysis (Cook 1977). Including outliers has no material impact on our results for U(ACSI).

Using the values in Table 6, the partial derivative of U(CASH) with respect to U(ACSI) for quadrant II observations (i.e., when QUAD_II = 1) is .018 ([.012 + .006 × 1]) and .010 for quadrant III observations ([.012−.002 (×1)]). For U(TOTALCASH), the equivalent figures are .020 for quadrant II observations and .014 for quadrant III observations.

We use Eq. 6 (i.e., we include both CLEAR SIGNAL and AMBIGUOUS SIGNAL observations) for the estimates reported in Panels A and B of Table 8. To maximise sample size, we include outliers in the sector and period estimates in Panel A. Four observations, which are outside of the SIC codes included in our analysis, are excluded from the sector tests. The estimates in Table 8 are for U(BONUS) as the dependent variable in all cases—the results when U(TOTALCASH) is the dependent variable are equivalent and are available from the authors on request.

One potential noteworthy aspect of these results is that one of our control variables (VAR(RET)) drops out of the analysis as time-invariant variables are dropped in fixed effects estimation. For this reason, we report our main results using Huber-White-Rogers estimation (Rogers 1993).

To investigate the potential impact of our decision to use lagged satisfaction, we examine an alternative measure of satisfaction—defined as the average of the contemporaneous and prior period’s measures of satisfaction relative to peer firms—as an alternative to U(ACSI) in Eq. 5. The coefficient estimates (p values) for satisfaction using this alternative measure for the regressions in columns (1) to (4) respectively of Table 5 are: [.019, (.009);.016 (.013);.009 (.036).009 (.023)]. These results show that the average measure—which combines both contemporaneous and lagged satisfaction relative to peer firms—is also positive and statistically significant.

The benefits of this estimation approach are discussed in depth in Gow et al. (2009).

All of the findings discussed here are confirmed for U(TOTALCASH) in unreported tests.

For these tests, U(ROA) is defined as the residual from the following regression: \( {\hbox{RO}}{{\hbox{A}}^{\rm{FIRM}}} = \alpha + {\beta_1}{\hbox{RO}}{{\hbox{A}}^{\rm{IND}}} + {\hbox{u}} \). Similarly, U(RET) is defined as the residual from the following regression: \( {\hbox{RE}}{{\hbox{T}}^{\rm{FIRM}}} = \alpha + {\beta_1}{\hbox{RE}}{{\hbox{T}}^{\rm{IND}}} + {\hbox{u}} \). U(ACSI) in these tests is defined as per Eq. 1.

A detailed discussion of the underlying rationale for the use of levels in such contexts is presented in Wooldridge (2006) and Fahlenbrach (2009).