Customer satisfaction and firm performance: insights from over a quarter century of empirical research

Another market measure of financial performance that has been the focus of investigation in the satisfaction-performance literature is Tobin’s q. However, concerns have been raised recently by Bendle and Butt (2018) regarding the use and misuse of Tobin’s q in marketing (see Bendle and Butt 2018). Because of these concerns (brought to our attention by a reviewer for which we are grateful, and a reviewer’s related suggestion), the analysis of Tobin’s q is presented in the Supplemental Material.

A breakout of the mean satisfaction-performance correlation by time period (i.e., year of study publication) similarly fails to illuminate a meaningful pattern of ever increasing or decreasing effect sizes. Specifically, for studies published (i) before 1997, M = −.110 (95% CI of −.224 to .007, n = 15), (ii) during the period 1998–2002, M = .092 (95% CI of .046 to .138, n = 42), (iii) during the period 2003–2007, M = .097 (95% CI of .064 to .131, n = 66), (iv) during the period 2008–2012, M = .140 (95% CI of .089 to .191, n = 90), and (v) during the period 2013–2017, M = .112 (95% CI of .044 to .180, n = 38). Thus, while the respective means for the last four time periods are each significantly stronger than the mean estimate for the time period “before 1997” (note, the mean for “before 1997” is not statistically significant), it should be noted that the respective means for each of the four, more recent time periods are not significantly different from one another.

Prior to estimating the final PM models, a full mediating (FM) model (i.e., marketing strategy, firm, and industry factors only directly impacting satisfaction and satisfaction then directly impacting performance) was estimated to determine whether fit would be superior. The MASEM findings indicate that the FM model is a poor fitting model across the performance measure while the respective partial-mediating versions of those models can be considered to fit the data well. Regarding the FM pooled performance model, the χ² (6) of 67.34 is significant (p < .05), the CFI of .57 and TFI of .04 are below the critical threshold of .95 suggestive of a good fit, the SRMR of .05 is reasonable, but values of zero represent perfect fit, and the RMSEA of .11 and its 90% CI having an upper bound that exceeds .10 (CI = .087 to .133) are further indicants of poor fit. Similarly, the FM model is a poor fit for market share. The Δχ² (5) of 921.56 is significant (p < .05), the CFI is .06 and TFI is −1.06 (which is a non-normed value), the SRMR is .16, and the RMSEA of .48 with a 90% CI of .455 to .508 are also indicants of poor fit. Likewise, the FM model is a poor fit to the data when profit is the performance measure. The χ² (6) of 36.83 is significant (p < .05), the CFI of .72 and TFI of .40 are below the threshold of .95, the SRMR of .04 is reasonable, but the RMSEA of .08 and its 90% CI having an upper bound that exceeds .10 (CI = .055 to .104) are indicative of poor model fit. The resulting modification indices (in the case of each of these three FM models) pointed to estimating a partial mediating model instead, with one or more of the exogenous factors further exerting a direct effect on the respective performance metric.