literature review

“Can You Hear Me Now?” Exit, Voice, and Loyalty under Increasing Competition T. Randolph Beard, Jeffrey T. Macher and John W. Mayo Auburn University

Georgetown University

Georgetown University

The authors gratefully acknowledge seminar participants at the Institutions and Innovation Conference at Harvard Business School and at the Federal Communications Commission (FCC) for helpful comments on earlier manuscript versions. The authors also received helpful comments from Chris Borek, Matthew Demartini, Silke Forbes, J. Bradford Jensen, Nathan Miller, Stanley Nollen, Karok Ray, Marcia Mintz, John Rust, Dennis Quinn, Rob Shapiro, Mike Stern, Scott Wallsten, and Luc Wathieu. The authors appreciate support from the Georgetown Center for Business and Public Policy in the McDonough School of Business; the Institute for Business Innovation at the University of California, Berkeley; and the Stanford Institute for Economic Policy Research. Any errors are attributable solely to the authors.

The Journal of Law and Economics

Vol. 58: , Issue. 3, : Pages. 717-745 (Issue publication date: August 2015)

https://doi.org/10.1086/684232

Abstract Competition works only if poorly performing vendors can be punished. The principal vehicle for consumers to discipline ill-performing firms is to switch to alternative providers. But switching is not the only mechanism consumers have to express disapproval. While some unhappy consumers may choose to no longer buy the good or service, other consumers express their disappointment through complaining. This article examines Albert O. Hirschman’s conjecture that as industries become more competitive, consumers’ complaints give way to switching. It offers a simple description of the theoretical relationships among market structure, quality, and complaints. It then utilizes an extensive data set to explore the empirical determinants of consumers’ complaining behavior in the local-exchange telephone industry. These data overcome the problem that complaints can depend on both competition and quality, while competition also presumably affects quality directly. The estimations accommodate this complication and provide considerable support for Hirschman’s conjecture.

[E]conomists have refused to consider that the discontented consumer might be anything but either dumbly faithful or outright traitorous (to the firm he used to do business with). (Hirschman , p. 31)

Notes The principal alternatives are disciplines

imposed by the threat or realization of regulatory or legal actions against a firm providing low quality.

For example, marketing research has principally focused on the role of consumers’ characteristics on the propensity to complain. See Kolodinsky ( ) for a review.

Our model assumes implicitly that failure has the same qualitative effect on each consumer. In particular, failure deprives consumers of some portion of the value of the service. As these values differ between types, the implications of failure are not the same for everyone. An alternative model parameterization would allow for differing degrees of failure, but this would not materially affect the conclusions if the value of functional service is held equal between customers. It would be necessary, however, to respecify the nonpurchase condition. Combining random values and random failure effects introduces substantial complexity without corresponding benefits to insight.

Our assumed ordering is predicated on two considerations. First, as a theoretical matter, a > b (in the presence of s > c) is required for switching behavior to be present at all. Second, the assumption generates the potential for an observed portfolio of consumer behaviors consistent with our empirical observations. See Section . However, one can easily imagine cases in which b > a. The parameterization assumed here is merely the simplest one that generates the observed portfolio of consumer behaviors.

We note that in his initial discussion of voice, Hirschman references not competition but monopoly, in which the cost of exit is prohibitive. In other words, the costs of voice and switching depend on market structure. Alternately, one could assume b > a, in which case c > s would be not only allowed but also necessary to generate all three observed behaviors.

We choose not to explore the possibility of a pure consumption value of complaining. We also assume that the benefits of complaining accrue solely to the complaining party.

The “quasi” modifier is necessitated by the prospect that changes in market structure and com-petition may generate changes in the quality—in turn, altering changes in propensities to complain. We account for this indirect effect in the empirical analysis.

For more information and data available via the Automated Reporting Management Information System (ARMIS), see FCC, ARMIS Data Descriptions (http://www.fcc.gov/encyclopedia/armis-data- descriptions-1).

While complaints are also made directly to the FCC, they represent a small fraction of the complaints received. We therefore utilize the more granular data afforded in the complaints made to local regulatory agencies.

See FCC, Local Telephone Competition and Broadband Deployment (http://www.fcc.gov/wcb/iatd/comp.html).

We account for this change in our empirical estimation.

Because individual competitive local- exchange carrier (CLEC) market shares are competitively sensitive, the FCC reports only aggregate CLEC market shares in each state- year. If only a small number of CLECs are present in a state-year (especially in 1999) the FCC withholds CLEC market-share data but provides data on the number of CLECs. In these instances, we first calculate average CLEC market share in a state-year once aggregate market-share data are reported (say, in 2000) using the data on number of CLECs. We then use this average CLEC market share to backcast aggregate CLEC market share in the unreported year using the average CLEC market share and the number of CLECs. This approach allows for a more complete data panel. As a robustness check, we estimated models with the smaller set of raw data provided by the FCC and confirm that no substantive differences obtain.

We experimented with several alternative specifications of the dependent variable, but no substantive changes to the empirical results reported below obtain.

In our measure, CLEC Countijt equals CLEC Countjt × (Linesijt/Max Linesjt), where Linesijt is the number of (residential or business) lines held by incumbent local-exchange carrier (LEC) i in state j at time t and Max Linesjt is the (residential or business) line count of the largest LEC in state j at time t. We also examined two alternative competition-count measures as robustness tests. We first substituted Sum Linesjt (the summation of lines in state j at time t) for Max Linesjt, which effectively proxies for the number of competitors an incumbent LEC faces by its share of total lines in that state-year. We then used CLEC Countjt (the logged statewide count of competitors). Neither of these alternative competition-count measures substantively alters the results or conclusions. For similar measurement approaches to competitive entry, see Abel ( ) and Greenstein and Mazzeo ( ), which use new-competitor counts, and Economides, Seim, and Viard ( ), which uses new-entrant market share.

Slightly less than 3 percent of the residential customer-satisfaction observations (19 of 668) are at 100 percent, while slightly more than 3 percent of the small-business customer-satisfaction observations (22 of 668) are at 100 percent. Accordingly, ceiling effects are not binding for the subsequent empirical analysis. We also experimented with different permutations (for example, the total of the aggregate satisfaction measure) as robustness checks. The results are substantially invariant to those reported.

Given the potential for correlations in the errors across our residential and business equations, we also estimated the residential and business models via seemingly unrelated regression as a robustness check. The results are inconsequentially different than those reported. We also explored whether nonlinearities from interactions or squaring terms added significantly to the power or insights of the model, but these alternatives were found to add little to the results reported here.

See Forbes ( ) for a similar result for the US airline industry.

Durbin-Wu-Hausman tests confirm endogeneity in the residential customer and business customer estimations, which suggests that an instrumental variables approach is warranted.

The specifics of these first-stage results, while not of primary interest, nonetheless are informative. For instance, we find that competition is negatively correlated with population for our measures of competition that are based on the number of firms in the market while being positively related to competition measures based on CLEC market share. This is consistent with the earlier entry by the largest CLECs (AT&T and MCI) that resulted in the capture of significant market shares, while smaller and more numerous CLECs targeted less populous states. For the count measure of competition, we also find, consistent with Abel (

), that the political orientation of public- utility commissions also impacted the number of entrants.

As a robustness check, we also created quality variables representing the summation of the percentage of satisfied consumers across the installation, repairs, and business-office categories. Using these variables in place of the average consumer-satisfaction variable produces similarly negative and statistically significant coefficient estimates in the residential and business models.

In separate regressions, we also sought to identify any direct influence of competition on our two quality measures. These regressions used our quality measures as a dependent variable, with both competition (number or market share) and a set of controls as independent variables. The resulting estimations provide considerable (but not unanimous) support for the intuitive proposition that increases in quality led to higher levels of satisfaction. In particular, the estimations indicate a positive and statistically significant effect from CLEC entry (CLEC CNTijt) on the percentage of residential customers reporting satisfaction (p <.01) and a positive and statistically significant effect from CLEC entry on the percentage of business customers reporting satisfaction (p <.01). Results for the estimations are more mixed when competition is measured by the market share of CLECs, with a statistically insignificant effect from CLEC market share on the percentage of residential customers reporting satisfaction but a positive and statistically significant effect from CLEC market share on the percentage of business customers reporting satisfaction (p <.05).

© 2015 by The University of Chicago Press. All rights reserved

View PDF (1767 KB)

View PDF PLUS (1860 KB)

1970

1.

2.

1995

3.

4.

3

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

2002

2006

2008

15.

16.

17. 2008

18.

19.

2002

20.

21.

SWIPE ACROSS ARTICLES�

1. Introduction

2. Theory

3. Empirics

4. Conclusion

� �

� � � � �

2018/10/29 00(15 第 1 ⻚页（共 1 ⻚页）