does providing competitive information to your...

35
Does Providing Competitive Information to Your Own Customers Increase Sales? by Guilherme Liberali Glen L. Urban and John R. Hauser May 2010 Guilherme (Gui) Liberali is a Visiting Scholar at MIT Sloan School of Management, and Assis- tant Professor of Marketing, Erasmus School of Economics, Erasmus University Rotterdam, 3000 DR Rotterdam, The Netherlands, (1.781) 632-7674, fax (+31)10-408-9169, libera- [email protected]. Glen L. Urban is the David Austin Professor of Marketing, MIT Sloan School of Management, Massachusetts Institute of Technology, E40-159, 1 Amherst Street, Cambridge, MA 02142, (617) 253-6615, [email protected]. John R. Hauser is the Kirin Professor of Marketing, MIT Sloan School of Management, Massa- chusetts Institute of Technology, E40-179, 1 Amherst Street, Cambridge, MA 02142, (617) 253- 2929, [email protected]. This research was supported by the MIT Sloan School of Management, the Center for Digital Business at MIT (ebusiness.mit.edu), and General Motors, Inc. We gratefully acknowledge the contributions of our industrial collaborators, research assistants, and faculty colleagues: Hunt Allcott, Eric Bradlow, Michael Braun, Luis Felipe Camargo, Daria Dryabura, Patricia Hawkins, Nick Pudar, Daniel Roesch, Dmitriy A Rogozhnikov, Joyce Salisbury, Catherine Tucker, and JuanJuan Zhang.

Upload: others

Post on 21-Aug-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Does Providing Competitive Information to Your Own Customers Increase Sales?

by

Guilherme Liberali

Glen L. Urban

and

John R. Hauser

May 2010 Guilherme (Gui) Liberali is a Visiting Scholar at MIT Sloan School of Management, and Assis-tant Professor of Marketing, Erasmus School of Economics, Erasmus University Rotterdam, 3000 DR Rotterdam, The Netherlands, (1.781) 632-7674, fax (+31)10-408-9169, [email protected]. Glen L. Urban is the David Austin Professor of Marketing, MIT Sloan School of Management, Massachusetts Institute of Technology, E40-159, 1 Amherst Street, Cambridge, MA 02142, (617) 253-6615, [email protected]. John R. Hauser is the Kirin Professor of Marketing, MIT Sloan School of Management, Massa-chusetts Institute of Technology, E40-179, 1 Amherst Street, Cambridge, MA 02142, (617) 253-2929, [email protected]. This research was supported by the MIT Sloan School of Management, the Center for Digital Business at MIT (ebusiness.mit.edu), and General Motors, Inc. We gratefully acknowledge the contributions of our industrial collaborators, research assistants, and faculty colleagues: Hunt Allcott, Eric Bradlow, Michael Braun, Luis Felipe Camargo, Daria Dryabura, Patricia Hawkins, Nick Pudar, Daniel Roesch, Dmitriy A Rogozhnikov, Joyce Salisbury, Catherine Tucker, and JuanJuan Zhang.

Page 2: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

1

Does Providing Competitive Information to Your Own Customers Increase Sales?

Abstract

A US automaker (“USAM”) had an asymmetric information problem: it had launched

excellent vehicles but consumers would not even consider USAM’s vehicles. To build trust and

encourage consideration and sales, USAM offered four types of competitive information: test

drives of competitive vehicles, competitive brochures, unbiased online purchase advisors, and

competitive community forums. USAM tested these strategies with a six-month 2 x 2 x 2 x 2

field experiment (year 1) and, to simulate a national launch, a six-month opt-in quasi-experiment

(year 2). To address conditional dependence of purchase on consideration, flow effects over mul-

tiple periods, continuous flows with discrete observations, and potential mediation by trust, we

use a (hidden) continuous-time Markov process which accounts for potential misclassifications

of consumers’ behavioral states. Competitive test drives and competitive brochures were identi-

fied as effective strategies for increasing consideration and purchase, but the effects were me-

diated through trust. We close with managerial implications.

Keywords: Asymmetric Information, Competitive Information, Continuous-time Markov

Processes, Communications, Electronic Marketing, Hidden States, Information

Search, Misclassification, Quasi-experiments, Trust

Page 3: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

2

INTRODUCTION AND MOTIVATION

Information is everywhere on the Internet. If a firm does not provide information to its

customers, someone else will. This is particularly true in the automotive market. Websites such

as Autotrader.com, Cars.com, ConsumerReports.org, Edmunds.com, Kelly Blue Book

(kbb.com), and TheAutoChannel.com compete to provide specifications, reviews, prices, and

availabilities for most makes and models. Automakers often do not control this information. On

the other hand, one of the most comprehensive sources of consumer information, test drives, are

offered only by manufacturer-affiliated dealers.

The study of competitive information is both theoretically and managerially interesting.

Theoretically, in a classic paper on “lemons,” Akerlof (1970, p. 500) argues that information

asymmetries are strategically important in markets “in which trust is important.” He argues that

new cars depreciate quickly because owners know which vehicles are “lemons” and which are

not, but sellers have no incentive nor credible mechanism to communicate that information. (To-

day certified pre-owned vehicle programs attempt to overcome the “lemons” problem by provid-

ing communication and seller commitment through warrantees.) In marketing, Urban (2004) ar-

gues that firms which provide unbiased competitive information to consumers will be rewarded

with trust which, in turn, leads to higher sales.

Information asymmetries and the need to generate trust became particularly acute in the

first decade of the 21st century. A US automaker (“USAM”) invested heavily in product devel-

opment and had launched vehicles that it believed were more reliable and satisfied consumer

needs significantly better than key competitors and USAM had evidence that consumers did not

share this belief. Drawing on experiences with USAM in the last 10-20 years consumers rejected

the automaker’s vehicles before searching for information on potential purchases. Because con-

Page 4: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

3

sumers never searched for information and never test drove USAM’s vehicles, those consumers

never bought USAM’s vehicles. Its sales suffered. Information asymmetries were a contributing

factor in its 2009 bankruptcy. For example, despite the facts that a USAM vehicle was tied with

Lexus for the top spot in J. D. Power’s 2007 vehicle dependability ranking, was the top US brand

in 2008 Consumer Reports, and was the number one brand in China, roughly half of all US con-

sumers (and almost 2/3rds in California) would not even consider USAM.

USAM’s Experiments (Year 1) and Quasi-Experiments (Year 2)

In this paper we analyze experiments and quasi-experiments by USAM to gain trust and

improve sales by providing competitive information to consumers. USAM provided consumers

with the ability to test drive 100 USAM and competitive vehicles, get unbiased competitive

eBrochures, have access to an unbiased web-based advisor that often recommended competitive

vehicles, and join an online community forum that discussed both USAM vehicles and competi-

tive vehicles. USAM wanted to test whether such competitive information would increase trust,

consideration, and purchase of USAM’s vehicles. (We believe the results are applicable beyond

the automotive market. Competitive information can be provided in many markets.)

Because these programs were strategically important, USAM evaluated potential effec-

tiveness and feasible over two years. In the first year USAM randomly assigned treatments in a 2

x 2 x 2 x 2 field experiment to identify which strategies have the potential to increase trust, con-

sideration, and sales. But treatments that are effective with forced exposure may not be cost-

efficient in a national roll-out. In the second year, USAM used a quasi-experiment (with opt-in)

to determine whether or not the competitive-information programs could be implemented na-

tionwide. To address opt-in self-selection in the second-year quasi-experiment, USAM used qua-

si-controls: (1) a control group in which consumers were not given opportunities for competitive

Page 5: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

4

information and (2) a forced-exposure group in which consumers were given strong incentives to

visit the website from which they could opt-in for competitive information.

Analysis Challenges

To evaluate USAM’s experiments and quasi-experiments we must address a number of

challenging issues. First, the impact of competitive information on sales might be indirect. Com-

petitive information might encourage consumers to consider USAM but not affect sales condi-

tioned on consideration. (This would indirectly increase sales because the pool of consumers

who consider USAM is now larger.) Competitive information might enhance consideration be-

cause consumers are more likely to consider a product if a firm lowers the cost of evaluating that

product relative to the brands already in the consideration set (Hauser and Wernerfelt 1990).

Alternatively, the impact of competitive information on consideration and purchase might

be mediated through trust. Mediation through trust is likely if competitive information addresses

information asymmetries by helping USAM to be perceived as a consumer advocate. There is

ample precedent in the literature for trust as a mediator of purchase or purchase intentions (e.g.,

Bart, et al. 2005; Büttner and Göritz 2008; Erdem and Swait 2004; Morgan and Hunt 1994; Por-

ter and Donthu 2008; Urban, Amyx, and Lorenzon 2009; Yoon 2002).

To model indirect effects our analyses are more complex than an average-effect compari-

son of test vs. control (although we do report such naïve analyses as a benchmark). First, to mod-

el indirect effects through consideration we account for the difficulty of directly measuring con-

sideration: empirical measures often misclassify “consideration” as “not considered,” or vice

versa. Second, automotive purchases occur over many months but competitive-information pro-

grams were available in specific months. For example, a competitive test drive might increase

trust in one month, enhance consideration (as measured imperfectly) the next month, and lead to

Page 6: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

5

a sale in yet another month. To analyze these dynamic data we use a “flow” model estimated

over the entire six-month period. Third, observed “flows” from “not consider” to “consider” and

to purchase (or not) may occur faster than the monthly interval of observation. We use conti-

nuous-time methods to infer the net effect of multiple unobserved paths to purchase even though

observations occur in discrete time.

Our primary method of analysis is a continuous-time Markov process (CTMP) with dis-

crete-time observation and subject to misclassifications of behavioral states. We begin by de-

scribing USAM’s year-1 experimental treatments and the measurements. We present naïve ana-

lyses and the one- and two-stage CTMP models. We address the impact of misclassification

analysis. We next describe the year-2 quasi-experiment, examine potential self-selection tests,

and compare analyses of the year-2 quasi-experiment to analyses of the year-1 experiment. The

results suggest that competitive test drives and competitive brochures enhance consideration and

sales, although the effects are mediated by trust. We close with a discussion of managerial impli-

cations.

YEAR-1 EXPERIMENTS: COMPETITIVE-INFORMATION STRATEGIES

The year-1 panel ran monthly from October 2003 to April 2004. (This was five years

prior to the bankruptcies of two US automakers.) Members of Harris Interactive’s panel were

screened to be in the market for a new vehicle in the next year, on average within the next 6.6

months, and invited to participate and complete six monthly questionnaires. In total, Harris Inter-

active enrolled 615 Los Angeles consumers of whom 317 completed all six questionnaires for an

average completion/retention rate of 51.5%. USAM did not retain recruitment rate statistics for

year 1, but, based on year 2, we estimate an initial recruitment rate of about 40%. Consumers

were assigned randomly to experimental cells in the 2 x 2 x 2 x 2 field experiment.

Page 7: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

6

Competitive Test Drives: Auto Show in Motion

Consumers were invited to an event at a California test track to drive vehicles from

BMW, Chrysler, Dodge, Ford, General Motors, Honda, Lexus, Mercedes, and Toyota without

sales pressure. In returned they received coupons good at Amazon.com and a chance to win

$10,000. Figure 1a reproduces information about the test drive, called Auto Show in Motion

(ASIM). ASIM had substantial fixed costs to set up the test track, procure and maintain over 100

vehicles, assure safety, and provide staff assistance. Per consumer costs, including recruiting, in-

centives, and lunch, were in the $50-100 range. ASIM was made available in period 4 to 39.1%

of the consumers.

Insert Figure 1 about here.

Customized Brochures

In year 1 this experimental treatment provided information about USAM rather than

competitive vehicles. (In year 2 USAM used competitive brochures.) Specifically, year-1 con-

sumers received brochures that were targeted to their specific needs as determined by measures

prior to the experiments (Figure 1b). The brochures were mailed in either period 2 or 3 to 51.7%

of the consumers.

Competitive Online Advisor: Auto Choice Advisor

Consumers were invited to use a web-based advisor that recommended vehicles based on

a series of questions that revealed the consumers’ wants and needs. The web-based advisor,

known as the Auto Choice Advisor (ACA), was similar to the advisor described in Urban and

Hauser (2004). See Figure 1c. ACA was made available in periods 2 through 6 to 49.2% of the

consumers. At the time of the experiments, USAM’s redesigned vehicles were just being

launched. ACA had a tendency to recommend Toyota (and other competitors) relative to USAM.

Page 8: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

7

Competitive Community Forum

Consumers participated in an CommuniSpaceTM online forum that discussed both USAM

and competitive vehicles. See Figure 1d. Consumers were free to participate in any of the over

30 dialogues which averaged over 30 comments in each. Consumers could discuss experiences

with their buying or ownership experience for any competitor and do so in an unbiased manner.

The community was made available in periods 2 through 5 to 47.6% of the consumers. At the

time of the experiment, due to consumers’ past experiences with USAM, USAM often did not

receive positive reviews in the forums.

Measures Used in the Analyses

Dependent measures. In each period consumers reported the vehicles that they were

“considering for your next purchase or lease.” They indicated the make-model combinations

from a drop-down menu of 348 make-model combinations. USAM was interested in whether or

not one of these vehicles was a USAM vehicle. With 348 make-model combinations on a drop-

down menu, misclassification was a real concern. The year-1 purchase dependent variable was

measured in the surveys. For modeling purposes, we assume that misclassification of the pur-

chase observation is small compared to misclassification of consideration.

Trust Mediator. Consumers have varied experiences in the auto market and varied per-

ceptions of USAM. Trust was central to USAM’s strategic thinking and, supported by the litera-

ture cited earlier, we believed that it was likely that trust would mediate the effect of competitive

information (and brochures) on consideration and purchase. Trust measures also enable us to

control for past history and to test for indirect effects. USAM measured consumers’ trust using a

five-item scale with items such as “I believe that this company is willing to assist and support

me.” or “Overall, this company has the ability to meet customer needs.” The items exhibited high

Page 9: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

8

reliability: Cronbach’s α = 0.95.

Covariates. Many automotive consumers are brand loyal. The first of six surveys col-

lected data on the vehicles consumers owned prior to the experiment. We dummy-code these da-

ta as “own USAM,” “own other American vehicle,” and “own Japanese.” The dummy variables

are relative to “own European.” USAM measured age with 11 categories, which we dummy-

code with 10 categorical variables. The covariates do not vary by observation period.

Treatment assignments. A few consumers could not access the treatment(s) to which

they were assigned. For example, some consumers experienced technical difficulties with ACA

and some could not come to ASIM. This appears to random and outside of experimental control.

To assess whether or not this randomness affected outcomes we compared analyses based on as-

signed treatment dummies to analyses based on self-reported treatment effects (e.g., did they vis-

it ASIM). Fortunately, the self-reports match up well with the assigned treatments and seem to

capture phenomena where consumers were not able to access the treatment. In comparative ana-

lyses the treatment-assignment coding and self-report coding provided the same strategic inter-

pretations. As an illustration, an online appendix compares trust regressions – there were no sig-

nificant differences in the estimated coefficients. To avoid redundancy in exposition, we report

only the analyses based on the dummy-coded self-reported treatment effects. We now describe

our analysis model and use it to analyze the year-1 experiment and the year-2 quasi-experiment.

CONTINUOUS TRANSITIONS AMONG BEHAVIORAL STATES

We are interested in whether or not competitive information, possibly mediated through

trust, encourages consumers to consider and/or purchase USAM vehicles. We represent this fo-

cus with the Markov diagram in Figure 2. The diagram is Markov because flows among beha-

vioral states depend only on the current behavioral state (and the explanatory variables), not the

Page 10: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

9

entire past history of flows.

Insert Figure 2 about here.

We index continuous time with t and we index behavioral states with i where i = 1, 2, and

3 index “do not consider USAM,” “consider USAM,” and “purchase USAM,” respectively. Let

1 if the consumer is in state i at time t, and let 0 otherwise. (To simplify notation

we suppress the subscript for consumer.) While consumers flow continuously among states and

may make multiple transitions in a given month, we only observe behavioral states at monthly

intervals. Let be the observation time for the monthly observation. Because we only ob-

serve the result of continuous transitions, we derive an expression for the probability, ,

that the consumer was in state i at and in state j at , where . Let be

the transition matrix of the ’s.

Competitive information and trust affect the rate at which consumers flow among states.

Let be the instantaneous flow rate during the observation period and let An be the flow

matrix of the ’s. Mathematically for j ≠ i, Δ is the probability that the consumer flows

from state i to state j in the time period between t and t + Δ for very small Δ during the ob-

servation period. For small Δt the only way to be in state j at time t + Δt is to be there at time t or

move there from another state in time Δ . This property gives the following differential flow eq-

uation:

(1) Δ 1 Δ Δ

Following Cox and Miller (1965) and Hauser and Wisniewski (1982) we let Δt → 0 to obtain a

differential equation for the transition matrix.

Page 11: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

10

(2)

The solution to the differential equation requires matrix exponentiation which is a difficult nu-

merical problem. Equation 3 provides the solution to the differential equation and provides two

ways by which the solution can be computed. ( is the matrix of Eigenvectors of and

exp Λ is the matrix with the exponentiation of the Eigenvalues on the diagonal.)

(3) !

exp Λ

To capture the effect of competitive information, trust, and the covariates we let the flows

be a function of these variables. By definition, the off-diagonal elements of are positive, so

we use logarithms. Let 1 if the consumer received experimental treatment k in observation

period n. Otherwise 0. Let be the trust as measured at the beginning observation pe-

riod n, that is, as measured at the end of the 1 period. Let ℓ be the ℓ covariate. We

represent flows by Equation 4 for the feasible behavioral-state transitions in Figure 2. We seek to

estimate the unknown parameters: , , , and ℓ .

(4) log ℓ ℓℓ

We code our dependent-variable observations such that 1 if the consumer is ob-

served in behavioral state i at and in behavioral state j at . Using Equations 3 and 4 we

write out the data likelihood. Although medical researchers have developed reversible jump

Monte Carlo Markov Chain estimation to obtain estimates of the parameters for moderately-

sized CTMP models (e.g., Suchard, Weiss and Sinsheimer 2001), matrix exponentiation presents

practical problems for many empirical applications, especially when the number of observed

transitions are small compared to the sample size as is the case in our data with USAM purchas-

Page 12: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

11

es. Sparse flows lead to numerical stability issues.1 We found maximum likelihood methods to

be more stable for the USAM experiments and quasi-experiments (and more common in the

CTMP literature). See Kulkarni (1995) for a review of computational methods to deal with ma-

trix exponentiation. (Our one-stage models have a running time of 1-5 hours with maximum-

likelihood estimation. Two-stage models are quicker.)

MODELING POTENTIAL MISCLASSIFICATION OF DEPENDENT MEASURES

Consideration is both transient and measured with potential error. For example suppose a

consumer is not quite sure whether he or she is seriously considering a USAM vehicle. The con-

sumer who is unsure might say he/she is considering a USAM vehicle in period 2 but not say so

in period 3. Even if the consumer is sure about consideration, he/she might miss a vehicle in a

drop-down menu of over 348 vehicles.2

Survey noise might cause us to infer phantom flows from “not consider USAM” to “con-

sider USAM” and back. These phantom flows might disguise the effects of the experimental

treatments, trust, or the covariates. To account for misclassification we define 1 if we

observe that the consumer says that he or she is in state i at the end of the observation period.

The true state, , is not observed. To simplify notation let and .

We seek to infer the probability of correct classification, Pr | , and of misclassifications,

Pr | for j ≠ i.

1 Pn is a stochastic matrix (rows sum to 1) which implies that the rows of An sum to zero. Thus the first Eigenvalue of Pn is 1.0 and describes steady-state behavior. The remaining Eigenvalues are less than 1.0 and describe the dy-namic behavior. If tn is too large or too small relative to the dynamic behavior of the system there will be only one non-zero Eigenvalue. The remaining Eigenvalues will be close to zero. Slight numerical errors could make them negative, which would imply imaginary flows when we use the logarithmic representation in Equation 3. This caus-es problems with a Bayesian sampler. 2 Toward the end of the decade USAM developed improved methods to measure consideration that mimicked the manner by which consumers choose consideration sets. See Hauser, et al. 2010.

Page 13: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

12

To model misclassification we adopt Jackson, et al.’s (2003) hidden Markov model. To

model misclassification we recognize that the likelihood of a series of observed states,

, , … , , is equal to the probability that we made those observations conditioned on a

set of true states, , , … , . If there were no restrictions, then for any observed state the

consumer could have been in any true state. For any sequence of observed states, if all transitions

were allowed, there is a non-zero probability for any of the 3x3x3x3x3x3 = 729 paths through

the true states at times , , … , . In our model “purchase” is a trapping state and assumed to

have negligible observation error, so the number of paths is far less, but not trivial. To form the

likelihood, we sum over all feasible true paths consistent with Figure 2. For example, the likelih-

ood that we observe “not consider” at the end of the first observation period and “consider” at

the end of the second observation period is given by Equation 5 where Pr , is a prior prob-

ability.

(5) Pr ,Pr | Pr Pr | Pr | Pr | Pr |

Pr | Pr Pr | Pr | Pr | Pr |

As we expand Equation 5 to all six periods the likelihood gets complicated, but is han-

dled easily by computer. Indeed, Jackson, et al. (2003, p. 197) provide a compact matrix notation

to sum the likelihood over all true paths. Based on the data likelihood we estimate the parameters

of the CTMP (the β’s) and the misclassification probabilities simultaneously.

TESTING THE DIRECT EFFECT OF COMPETITIVE INFORMATION

Average-Effect Analyses

We begin with average-effect analyses. At the end period 4 among consumers who expe-

rienced ASIM in period 4, 60% said they would consider USAM. This is significantly more than

the 41% measured among consumers who did not experience ASIM (t = 3.4, p < .01). ASIM also

Page 14: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

13

had a significant effect on purchase (12% vs. 4%, t = 2.9, p = .01. Average-effect analyses do not

identify significant treatment effects for customized brochures, the competitive online advisor, or

the competitive community forum, all of which occur over multiple periods. However, these

naïve analyses do not address the conditional dependence of purchase on consideration, media-

tion by trust, flow effects over multiple periods, continuous flows with discrete observations, and

misclassification. These complications also make multi-period ANOVAs exceeding hard to in-

terpret. Instead, we turn to the more-complete CTMP model.

Continuous-Time Markov Process Model with Potential Misclassification

Table 1 summarizes direct estimation with the CTMP model. To simplify Table 1, we do

not report ’s for the covariates – none were significant at the 0.10 level. (Details available from

the authors.) Because of the logarithmic specification, a negative coefficient indicates that a flow

rate decreased, not that the flow is negative. Misclassification was moderate; approximately 12%

of the consumers were estimated to be misclassified as “consider USAM” when they did not yet

consider USAM and approximately 6% were estimated to be misclassified as “do not consider

USAM” when they considered USAM.

Insert Table 1 about here.

Each of the four sets of two columns represents one of the four allowable flows in Figure

2. Table 1 indicates that there is no identifiable direct effect due to the competitive information

(the experimental treatments). However, lagged trust significantly increases flows from consid-

eration to purchase and significantly decreases flows from “consider” to “do not consider.” We

examine next whether the effect of competitive information is mediated through trust.

To anticipate a two-stage model, we estimated a reduced-form CTMP model with only

lagged trust as a variable. The results, shown in the lower portion of Table 1, have similar impli-

Page 15: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

14

cations: trust is a key driver of purchase and of maintaining consideration.

COMPETITIVE INFORMATION ACTS THROUGH TRUST

To examine the mediating effects of trust we develop a two-stage model. For the first

stage we regress trust on competitive test drives, customized brochures, competitive advisors,

and competitive community forums. We include lagged trust because the experimental treat-

ments are likely to increase or decrease trust and we include covariates to account for unob-

served propensity for trusting USAM. We include dummy variables for observation periods to

account for unobserved advertising and other actions by USAM and to account for unobserved

environmental shocks. (Period 1 is a pre-measure and the period-2 dummy variable is set to zero

for identification.) The period dummy variables also account for any measurement artifact that

might boost trust (“Hawthorne” effect). The trust regression is summarized in Table 2. (To sim-

plify exposition we suppress the categorical age variables; none were significant.)

Insert Table 2 about here.

Table 2 suggests that both competitive test drives (ASIM) and customized brochures in-

crease consumers’ trust in USAM. Given the importance of trust in the CTMP stage and given

that there is no direct effect for ASIM and brochures in a one-stage model, this is an important

finding. The significance of ASIM is consistent with naïve analyses. The impact of customized

brochures is consistent with other published studies of customization (e.g., Ansari and Mela

2003; Hauser, et al. 2009). We believe that such field-experimental evidence of the positive ef-

fect of competitive information is relatively novel.

To examine the second stage of the CTMP analysis, we use estimated trust ( ̂ ) from

the first stage rather than measured trust ( ) in the CTMP conditional likelihood. These two-

stage estimates are limited-information maximum-likelihood (LIML) estimates. LIML estimates

Page 16: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

15

are consistent but we need to use bootstrap methods to check the standard errors for the ’s

(Berndt, et al. 1974; Efron and Tibshirani 1994; Wooldridge 2002, p. 354, 414). Table 3 reports

both the LIML estimates and the means from 1,000 bootstrap replicates.

Insert Table 3 about here.

The estimates in Table 3, both LIML and bootstrap, are similar to the reduced model in

Table 1 confirming the impact of trust on consideration and purchase.3 The two-stage model con-

firms that some competitive information (competitive test drives) increases consideration and

purchase, but that the effect is through increased trust. This finding suggests that USAM explore

further strategies to increase trust. We discuss such strategies later in the paper.

We did not find an effect for the competitive advisor and the competitive forum. This was

not unexpected. As described in an earlier section, both ACA and community forums provided

competitive information that did not favor USAM. While a willingness to provide competitive

information might enhance trust, negative recommendations from a online advisor (ACA) and

community forums appear to have offset USAM’s gesture.

THE IMPACT OF MODELING MISCLASSIFICATION IN A CTMP MODEL

Although misclassification in CTMP is relatively new to marketing, it has been used suc-

cessfully in the diagnosis of disease progression, the diagnosis of toxoplasmosis infection, the

spread of HIV, the probability of workers changing jobs, volcanic activity, and Monte Carlo stu-

dies of forecasting error (Aspinall, et al. 2005; Bessec and Bouabdallah 2005; Chen and Sen

3 The LIML and reduced-form estimates are almost identical. The LIML and bootstrap coefficients of lagged trust are not significantly different. The constants representing flows between “not consider USAM” and “consider USAM” do differ, but offset. This difference is likely due to misclassification analysis interacting with the bootstrap procedure. Specifically, because bootstrap randomly selects observations, many observations are repeated in a given replicate causing that replicate to underestimate misclassification. The means of the 1,000 replicates were almost identical to the medians suggesting that outliers were not a problem. The running time is about 250 hours for 1,000 two-stage replicates.

Page 17: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

16

2007; Hausman, Abrevaya and Scott-Morton 1998; Jackson, et al. 2003; Rahme, Joseph and

Gyorkos 2000; Tzavidis and Lin 2006). We test the impact of modeling misclassification in

USAM’s experiment by using 5-fold cross-validation to examine predictive ability. For each of

five subsamples we estimate the two-stage CTMP model with 80% of the data and predict beha-

vior for the remaining 20%. Accounting for misclassification improves cross-validation hit rates

significantly from 66.1% to 77.1% (p < 0.001). The coefficients themselves also vary. When we

ignore misclassification, the constants and lagged (estimated) trust remain significant in the same

pattern as in Table 3, but we underestimate the impact of trust for maintaining consideration.

YEAR 2 QUASI-EXPERIMENTS ON COMPETITIVE INFORMATION

Buoyed with the success of the year-1 experiment USAM sought to test the feasibility of

a national launch of competitive information. The year-2 quasi-experiment analyzed competitive

test drives (ASIM) but expanded the competitive information treatments to include competitive

brochures. USAM maintained both the competitive advisor and the community forum even

though they had no identifiable effect in year 1.

In year 2 USAM simulated a national launch of competitive-information strategies by al-

lowing consumers to opt-in to the strategies. (For example, ASIM costs of $50-100 per untar-

geted consumer were not feasible on a national basis. Opt-in targeting is necessary to make the

costs reasonable.) By its very nature, opt-in makes USAM’s test a quasi-experiment requiring us

to examine potential self-selection. USAM included two quasi-controls. Consumers were as-

signed randomly to one of three cells. Consumers in the control cell received no treatments. Con-

sumers in the forced-exposure cell were invited to an “Internet study” that included a visit to

USAM’s “My Auto Advocate” website at which they could opt-in to competitive information

treatments. See Figure 3a. Consumers in the pure opt-in cell received an advertisement inviting

Page 18: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

17

them visit the “My Auto Advocate” website.

Insert Figure 3 about here.

The year-2 panel ran monthly from January to June, 2005. Members of Harris Interac-

tive’s panel were again screened to be in the market for a new vehicle, on average within the

next 2.2 years, and invited to participate and complete six monthly questionnaires. (The year-1

sample was restricted to 12-month intenders, the year-2 sample was not.) Incentives were similar

to year 1. In total, Harris Interactive invited 6,092 Los Angeles consumers of which 1,720 com-

pleted all six questionnaires for an average response/completion/retention rate of 28.2%. This

rate was not significantly different across the three conditions (p = .25).

Competitive Test Drives: Auto Show in Motion

Consumers were invited to an event at one of three test tracks. Otherwise ASIM was sim-

ilar to year 1. Respondents received 20 reward certificates (worth $1 each) for participating.

Competitive Brochures and USAM Booklets

In year 2 USAM continued to offer USAM brochures (called eBooklets), but this time on

an opt-in basis. To test competitive information, consumers could also download competitive

brochures. Although many competitive brochures were available on manufacturers’ websites,

USAM’s single-source webpage made it more convenient for consumers to compare vehicles.

Consumers received 5 reward certificates for downloading a USAM brochure. See Figure 3b.

Competitive Online Advisor: My Product Advisor

USAM updated ACA to “My Product Advisor (MPA)” and made it available directly

from the “My Auto Advocate” website. Besides an improved interface, MPA had a “garage” at

which consumers could store vehicle descriptions. Like ACA, MPA was unbiased. Consumers

received 5 reward certificates for using MPA. See Figure 3c.

Page 19: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

18

Competitive Community Forum

The community forum was updated and integrated with the “My Auto Advocate” web-

site. Consumers received at least 5 reward certificates for participating, but could earn up to 20

reward certificates for active participation. See Figure 1d.

Trust, Dependent Measures, and Covariates

Trust, consideration, and purchase were measured as in year 1, as were prior ownership

of USAM, other American, and Japanese vehicles. Age was measured directly rather than by

categories and sex of the respondent was recorded.

EXAMINING ISSUES OF SELF-SELECTION TO TREATMENTS

It is possible that consumers who take advantage of competitive information are more

likely to be interested in purchasing an automobile in the near future, but this self-selection may

or may not translate into greater consideration for or purchase of USAM vehicles.

The Effect of Cell Assignment for Non-Participating Consumers

Many consumers in the opt-in (35.9%) and forced-exposure (45.0%) cells and, by defini-

tion, all consumers (100%) in the control cell did not opt-in to treatments. If self-selection im-

pacts the dependent measures, then we should have a non-random removal of consumers from

the opt-in and forced-exposure cells. If self-selection compromises the quasi-experiment these

consumers should be less likely to consider and/or purchase USAM vehicles. To analyze this ef-

fect we created dummy variables for experimental cell and estimated a CTMP model with mis-

classification. There are no significant effects due to cell assignment (all p’s > 0.14 for all dum-

mies for all flows; often much higher). We obtain similar results when we include trust.

Exposure to My Auto Advocate

We have already established that the there is no significant difference in net response rate

Page 20: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

19

between those consumers who (a) were required to visit “My Auto Advocate” and completed the

six questionnaires (forced exposure) and (b) those who could opt-in to “My Auto Advocate” and

completed the six questionnaires (p > 0.17). To examine whether “My Auto Advocate” itself had

a significant effect on consideration and purchase we repeat the dummy-variable CTMP model

with misclassification, but for all consumers. The cell-assignment dummies remain insignificant

(all p’s > 0.23) with and without including trust in the model.

While these tests do not unequivocally rule out self-selection, they suggest it is reasona-

ble to examine the potential impact of providing consumers with competitive information and to

compare the effects to the year-1 experiment.

YEAR-2 ANALYSIS OF COMPETITIVE INFORMATION

We repeated all year-1 analyses with the year-2 data. As in year 1, none of the quasi-

experimental treatments had direct significant effects in the CTMP model and the constants and

coefficients of lagged trust in the reduced model were similar to those in the full model. Misclas-

sification was slightly lower than in year 1; approximately 9% of the consumers were estimated

to be misclassified as “consider USAM” when they did not yet consider USAM and approx-

imately 3% were estimated to be misclassified as “do not consider USAM” when they actually

considered USAM. Accounting for misclassification increased the predictive ability in a five-

fold cross-validation from 74.7% to 87.1% (p < 0.001). Because the two-stage LIML and boot-

strap estimates are similar to the reduced one-stage model we simplify exposition by reporting

only the two-stage estimates in Tables 4 and 5.4 Other details are available from the authors.

Insert Tables 4 and 5 about here.

4 All reduced-model and LIML estimates did not differ significantly (all p’s > .32). Differences in the LIML and bootstrap constants are explained in footnote 3. The mean and median bootstrap estimates are almost identical. One coefficient of lagged trust was significant for the bootstrap estimates, but not for the LIML estimates, however, the value of the coefficient itself is almost identical (0.468 vs. 0.474).

Page 21: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

20

In year 1 both lagged trust and estimated lagged trust had significant effects on considera-

tion and purchase (e.g., p ≤ 0.03, bootstrap estimates). In year 2 both lagged trust and estimated

lagged trust also have significant effects on consideration and purchase (p ≤ 0.05, bootstrap esti-

mates). (The bootstrap standard errors are more accurate than the LIML standard errors.) In-

creasing trust is clearly beneficial for USAM in year 2 as it was in year 1.

Competitive test drives still increase trust in year 2, although the effect is now only mar-

ginally significant (p = 0.10). The interesting new implication is that competitive brochures in-

creased trust in USAM significantly in year 2 (p = 0.04). Sex and age are significant in year 2

possibly reflecting unobserved changes in USAM’s product mix or advertising. Overall, the

year-2 two-stage model is remarkably similar to the year-1 model despite changes in the experi-

mental treatments, unobserved changes in the environment, the change from an experiment to a

quasi-experiment, and a sample that is not limited to consumers who plan to purchase in 12

months. Tables 1 through 5 suggest strongly that competitive information is an effective strategy

to increase trust (in USAM) and, through trust, to increase consideration and purchase of USAM

vehicles.

These quantitative conclusions are consistent with qualitative comments by consumers

who participated in the quasi-experiment:

My Auto Advocate: “I've learned a lot more about USAM. I didn't realize how many dif-

ferent models of cars they own. It was eye opening. I enjoyed it and I have a more positive view

of USAM than previously.”

Competitive test drives: “Please don't stop doing these events, USAM. This was the one

and only reason we purchased a USAM car over a Mustang GT or a Dodge Magnum R/T. There

was no way we would have test-driven a USAM car had it not been for ASIM. It's the best expe-

rience I can imagine for overcoming people's prejudices against USAM and selling them on your

Page 22: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

21

many terrific products (like the …). I doubt I'd make the trip to different dealers (…) to drive

those models.”

Competitive test drives: “I liked the show because I was able to test drive many differ-

ent makes and models of cars all at one time. It also opened my eyes to some cars which I may

not have originally considered.”

Competitive information on the website: “The competitive specs were good for com-

paring similar specs to the one chosen on the proving grounds. The testimonials were good to

hear from others who tested the car and were normal people instead of dealers. I like how it let

you choose the vehicle by style or brand….”

MANAGERIAL IMPLICATIONS

US automobile manufacturers face challenges as they try to improve sales and profit af-

ter emerging from bankruptcy. Our analyses have two implications. First, trust is key. Those

consumers who trust USAM are more likely to consider and purchase USAM vehicles. Second,

competitive test drives and competitive product brochures influence consumers to avoid rejecting

USAM vehicles before they gather information. The effect is predominately through increased

trust in USAM. USAM’s managers felt the results were consistent with qualitative data and suf-

ficiently compelling to investigate further. At minimum USAM’s experiments and quasi-

experiments provide evidence that honest competitive comparisons are effective for firms that

are disadvantaged by low consideration of their products despite having good products.

Competitive test drives and competitive brochures were not rolled out nationally because

of other distractions during the automotive and financial crises at the end of the first decade of

the 21st century. Paying for competitive test drives in a large traveling ASIM format is expensive

and was not judged to be cost effective, but other options were explored by USAM. For example,

a test program at a USAM dealer in Phoenix proved cost-effective for an SUV competitive test

drive. USAM continues to experiment with competitive test drives in key local markets to identi-

Page 23: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

22

fy formats with the best benefit-to-cost ratio. Competitive brochures are clearly attractive strate-

gies, especially with the growth of computer power and Internet use. USAM now uses competi-

tive comparisons on its website in a capability called “May the Best Car Win.” The initiative

uses standardized Polk data on prices, specifications, and equipment for preselected competition

and consumer-specified vehicles. One of us recently visited a wide variety of dealers to test drive

small SUVs. Only USAM dealers offered unsolicited extended weekend test drives and encour-

aged competitive comparisons.

USAM also implemented communications strategies. In 2009 a series of advertisements

starring a well-known sports announcer made explicit comparisons to competitors featuring

“surprises” such as good fuel economy. In the fall of 2009 USAM implemented a policy by

which consumers could try vehicles for 60 days and return them if they were not satisfied (in ef-

fect, an extended test drive). All of these marketing tactics sought to encourage competitive

comparisons and/or increase trust in USAM.

SUMMARY

If a firm has products that are much better than consumers perceive them to be (asymme-

tric information), USAM’s experiments and quasi-experiments suggest that consideration and

purchase increase if the firm makes competitive information available. (If a firm has inferior

products then providing competitive information to consumers may not be profitable as sug-

gested by the insignificant effects for ACA and the online-community.) The effect of competitive

information appears to have been mediated through trust. Providing competitive information

likely led consumers to place more trust in USAM. In turn, trust increased demand through con-

sideration and purchase of USAM’s vehicles. Managerially, the open question is whether provid-

ing competitive information has an attractive benefit-to-cost ratio. Local competitive test drives,

Page 24: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

23

providing competitive brochures, and competitive communications appear to be more cost effec-

tive than a national ASIM, but all of these tactics await formal experimental tests.

To analyze the experiments and quasi-experiments we used CTMP models to address the

data constraint that transitions occur continuously while we only observe the results of these

transitions at monthly intervals. While CTMP and related models, including hidden Markov

models, have been used successfully to analyze other marketing issues, we are unaware of other

applications in marketing that address potential misclassification errors in CTMPs (Ding and

Eliashberg 2008; Eliashberg, et al. 2000; Hauser and Wisniewski 1982; Netzer, Lattin and Srini-

vasan 2008; Roberts, Morrison and Nelson 2004, 2005; Srinivasan and Kim 2009; Weerahandi

and Moitra 1995). The analysis of misclassification proved to be a key methodological compo-

nent of our analyses.

CTMP, accounting for misclassification, is a powerful tool. Because flows were sparse in

the USAM experiments and quasi-experiments, stability issues made maximum likelihood esti-

mation the best numeric technique. When flows are less sparse, matrix exponentiation is more

stable numerically and MCMC analysis will be computationally feasible. When there is more da-

ta per consumer than in the USAM experiments and quasi-experiments, CTMP analyses could be

extended to include heterogeneity in flow-rate or trust-regression coefficients.

Page 25: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

24

REFERENCES

Ansari, Asim and Carl F. Mela (2003), “E-Customization,” Journal of Marketing Research, 40,

(May), 131-145.

Aspinall, W. P., R. Carniel, O. Jaquet, G. Woo, and T. Hincks (2006), “Using Hidden Multi-

State Markov Models with Multi-parameter Volcanic Data to Provide Empirical Evi-

dence for Alert Level Decision-Support,” Journal of Volcanology and Geothermal Re-

search, 153, 112-124.

Bart, Yakov, Venkatesh Shankar, Fareena Sultan, and Glen L. Urban (2005), “Are the Drivers

and Role of Online Trust the Same for All Web Sites and Consumers? A Large-Scale

Exploratory Empirical Study,” Journal of Marketing, 69, (October), 133-152.

Berndt, Ernie K., B. H. Hall, R. E. Hall, and Jerry Hausman (1974), “Estimation and Inference in

Nonlinear Structural Models,” Annals of Social Measurement, 3, 653-665.

Bessec, Marie and Othman Bouabdallah (2005), “What Causes The Forecasting Failure of Mar-

kov-Switching Models? A Monte Carlo Study,” Studies in Nonlinear Dynamics & Eco-

nometrics, 9, 2, Article 6.

Büttner, Oliver B. and Anja S. Göritz (2008), “Perceived Trustworthiness of Online Shops,”

Journal of Consumer Behavior, 7 (1), 35–50.

Chen, Pai-Lien and Pranab K. Sen (2007), “Markov Chain Model Selection by Misclassified

Model Probabilities,” Communications in Statistics—Theory and Methods, 36, 143–153.

Cox, David R. and Hilton D. Miller (1965), The Theory of Stochastic Processes, (London, UK:

Chapman & Hall).

Ding, Min and Jehoshua Eliashberg (2008), “A Dynamic Competitive Forecasting Model Incor-

porating Dyadic Decision Making,” Management Science, 54, 4, 820-834.

Efron, Bradley and Robert J. Tibshirani (1994), An Introduction to the Bootstrap, (New York,

Page 26: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

25

NY: Chapman & Hall/CRC).

Eliashberg, Jehoshua, Jedid-Jah Jonker, Mohanbir S. Sawhney, and Berend Wierenga (2000),

Marketing Science, 19, 3, 226-243.

Erdem, Tülin and Joffre Swait (2004), “Brand Credibility, Brand Consideration, and Choice,”

Journal of Consumer Research, 31 (June), 191-98.

Hausman, Jerry A., Jason Abrevaya and Fiona M. Scott-Morton (1998), “Misclassification of the

Dependent Variable in a Discrete-response Setting,” Journal of Econometrics, 87, 239-

269.

Hauser, John R., Olivier Toubia, Theodoros Evgeniou, Daria Dzyabura, and Rene Befurt (2010),

“Cognitive Simplicity and Consideration Sets,” forthcoming Journal of Marketing Re-

search, 47, (June).

------, Glen L. Urban, Guilherme Liberali, and Michael Braun (2009), “Website Morphing,”

Marketing Science, 28, 2, (March-April), 202-224.

------ and Birger Wernerfelt (1990), "An Evaluation Cost Model of Consideration Sets," Journal

of Consumer Research, Vol. 16, (March), 393-408.

------ and Wisniewski, Kenneth J. (1982), “Dynamic Analysis of Consumer Response to Market-

ing Strategies,” Management Science, 28, 5, 455-484.

Jackson, Christopher H., Linda D. Sharples, Simon G. Thompson, Stephen W. Duffy and Elisa-

beth Couto (2003), “Multistate Markov Models for Disease Progression with Classifica-

tion Error,” The Statistician, 52, 193-209.

Kulkarni, V. (1995), Modeling and Analysis of Stochastic Systems, (London, UK: Chapman &

Hall/CRC).

Morgan, Robert M. and Shelby D. Hunt (1994), “The Commitment—Trust Theory of Relation-

Page 27: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

26

ship Marketing,” Journal of Marketing, 58 (3), 20–38.

Netzer, Oded, James M. Lattin and V. Srinivasan (2008),” A Hidden Markov Model of Customer

Relationship Dynamics,” Marketing Science, 27, 2, 185-204.

Porter, Constance Elise and Naveen Donthu (2008), “Cultivating Trust and Harvesting Value in

Virtual Communities,” Management Science, 54, 1, (January), 113-128.

Rahme, Elham, Lawrence Joseph, and Theresa W. Gyorkos (2000), Applied Statistics, 49, 119-

128.

Roberts, John H., Pamela D. Morrison and Charles J. Nelson (2004), “Implementing A Pre-

launch Diffusion Model: Measurement And Management Challenges Of The Telstra

Switching Study,” Marketing Science, 23, 2, 186-191.

-----, -----, and ----- (2005), “A Prelaunch Diffusion Model for Evaluating Market Defense Strat-

egies,” Marketing Science, 24, 1, 150-164.

Srinivasan, V. and Sang-Hoon Kim (2009), “A Conjoint-Hazard Model of the Timing of Buyers'

Upgrading to Improved Versions of High-Technology Products,” Journal of Product In-

novation Management, 26, 3, 278-290.

Suchard, Marc A., Robert E. Weiss and Janet S. Sinsheimer (2001), “Bayesian Selection of Con-

tinuous-Time Markov Chain Evolutionary Models,” Molecular Biology Evol., 18, 6,

1001-1013.

Tzavidis, Nikos and Yan-Xia Lin (2006), “Estimating from Cross-Sectional Categorical Data

Subject to Misclassification and Double-Sampling: Moment-based, Maximum Likelihood

and Quasi-Likelihood Approaches,” Journal of Applied Mathematics and Decision

Sciences, 2006, 1-13.

Urban, Glen L. (2004), “The Emerging Era of Customer Advocacy,” MIT Sloan Management

Page 28: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Competitive Information to Your Own Customers

27

Review, (Winter), 45, 2, 77-82.

------, Cinda Amyx, and Antonio Lorenzon (2009), “Online Trust: State of the Art, New Fron-

tiers, and Research Potential,” Journal of Interactive Marketing, 23, 179-190.

------ and John R. Hauser (2004), “’Listening-In’ to Find and Explore New Combinations of Cus-

tomer Needs,” Journal of Marketing, 68, (April), 72-87.

Weerahandi, Samaradasu and Soumyo Moitra (1995), “Using Survey Data to Predict Adoption

and Switching for Services,” Journal of Marketing Research, 32, 1, 85-96.

Wooldridge, Jeffrey M. (2002), “Econometric Analysis of Cross Section and Panel Data,” (Cam-

bridge, MA: MIT Press).

Yoon, Sung-Joon (2002), “The Antecedents and Consequences of Trust in Online Purchase De-

cisions,” Journal of Interactive Marketing, 16 (2), 47–63.

Page 29: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

TABLE 1 ONE-STAGE ANALYSIS OF COMPETITIVE INFORMATION

CTMP with Misclassification Modeled

State at Tn-1: From “Do Not Consider USAM” From “Consider USAM”

State at Tn: to Consider USAM to Purchase USAM to Do Not Consider USAM to Purchase USAM

Full Model p value p value p value p value

Constant 0.005 b 0.10 0.022 b 0.10 0.013 b 0.09 0.084 a 0.01

Lagged Trust -0.043 c 0.84 0.607 c 0.42 -0.470 b 0.10 0.278 a 0.04

Competitive Test Drives

-0.071 c 0.97 -2.138 c 0.86 -0.058 c 0.94 -0.359 c 0.50

Customized Brochures

0.747 c 0.28 1.639 c 0.25 -4.832 c 0.60 0.189 c 0.53

Competitive Online Advisor

-1.130 c 0.21 6.075 c 0.55 -0.639 c 0.47 -0.138 c 0.64

Competitive Forum

0.365 c 0.57 0.288 c 0.85 0.767 c 0.35 0.185 c 0.53

Reduced Model

Constant 0.043 a <0.01 0.001 b 0.10 0.066 a <0.01 0.121 a <0.01

Lagged Trust 0.06 c 0.75 0.556 c 0.63 -0.502 a 0.01 0.255 a 0.04

a Significant at the 0.05 level (shown in bold). b Significant at the 0.10 level, but not 0.05 level (shown in bold italics)

Page 30: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

TABLE 2 TRUST AS A FUNCTION OF COMPETITIVE INFORMATION (YEAR 1)

Key Variables Covariates Effect p value Effect p value

Intercept 0.640 b 0.06 Own USAM 0.000 c 0.84

Lagged Trust 0.860 a <0.01 Own American 0.021 c 0.59

Competitive Test Drives 0.380 a <0.01 Own Japanese -0.019 c 0.62

Customized Brochures 0.171 a <0.01 Period 3 -0.220 a <0.01

Competitive Advisor -0.057 c 0.39 Period 4 -0.295 a <0.01

Competitive Forum 0.045 c 0.22 Period 5 -0.127 a <0.01

Adjusted R2 0.749 c Period 6 -0.251 a <0.01

a Significant at the 0.05 level (shown in bold). b Significant at the 0.10 level, but not 0.05 level (shown in bold italics)

TABLE 3 SECOND-STAGE ANALYSIS BASED ON ESTIMATED LAGGED-TRUST (YEAR 1)

CTMP with Misclassification Modeled

State at Tn-1: From “Do Not Consider USAM” From “Consider USAM”

State at Tn: to Consider USAM to Purchase USAM to Do Not Consider USAM to Purchase USAM

LIML p value p value p value p value

Constant 0.042 a <0.01 0.001 b 0.08 0.066 a <0.01 0.120 a <0.01

Lagged Trust (estimated)

0.123 c 0.60 -0.380 c 0.65 -0.471 b 0.06 0.275 b 0.06

Bootstrap (1,000 replicates)

Constant 0.100 a <0.01 0.001 c 0.36 0.156 a <0.01 0.129 a <0.01

Lagged Trust (estimated)

0.209 c 0.13 -0.130 c 0.69 -0.325 a 0.03 0.275 a 0.01

a Significant at the 0.05 level (shown in bold). b Significant at the 0.10 level, but not 0.05 level (shown in bold italics)

Page 31: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

TABLE 4 TRUST AS A FUNCTION OF COMPETITIVE INFORMATION (YEAR 2)

Key Variables Covariates Effect p value Effect p value

Intercept 0.902 a <0.01 Own USAM 0.132 c <0.01

Lagged Trust 0.824 a <0.01 Own American 0.023 c 0.19

Competitive Test Drives 0.089 b 0.10 Own Japanese -0.059 c <0.01

Competitive Brochures 0.050 b 0.04 Sex 0.037 a 0.02

USAM Booklets 0.017 c 0.52 Age -0.088 a 0.02

Competitive Advisor -0.020 c 0.39 Age-Sq/100 0.009 a 0.01

Competitive Forum 0.003 c 0.91 Period 3 0.038 c 0.15

Period 4 0.040 c 0.13

Period 5 0.028 c 0.30

Adjusted R2 0.700 c Period 6 0.050 b 0.06

a Significant at the 0.05 level (shown in bold). b Significant at the 0.10 level, but not 0.05 level (shown in bold italics)

TABLE 5 SECOND-STAGE ANALYSIS BASED ON ESTIMATED LAGGED-TRUST (YEAR 2)

CTMP with Misclassification Modeled

State at Tn-1: From “Do Not Consider USAM” From “Consider USAM”

State at Tn: to Consider USAM to Purchase USAM to Do Not Consider USAM to Purchase USAM

LIML p value p value p value p value

Constant 0.025 a <0.01 0.001 a <0.01 0.099 a <0.01 0.006 a <0.01

Lagged Trust (estimated)

0.989 a <0.01 0.468 c 0.36 -0.425 a <0.01 -0.018 c 0.96

Bootstrap (1,000 replicates)

Constant 0.072 a <0.01 0.001 a 0.01 0.188 a <0.01 0.006 a <0.01

Lagged Trust (estimated)

0.614 a <0.01 0.474 a 0.05 -0.362 a <0.01 0.060 c 0.87

a Significant at the 0.05 level (shown in bold). b Significant at the 0.10 level, but not 0.05 level (shown in bold italics)

Page 32: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

FIGURE 1 YEAR-1 COMPETITIVE-INFORMATION EXPERIMENTAL TREATMENTS

(a) Competitive Test Drive (b) Customized Brochures

(c) Competitive Online Advisor (d) Competitive Community Forum

Site contained over 60

dialogues averaging over 60 comments.

Page 33: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

FIGURE 2 CONTINUOUS-TIME FLOWS AMONG BEHAVIORAL STATES

Page 34: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

FIGURE 3 YEAR-2 COMPETITIVE-INFORMATION QUASI-EXPERIMENTAL TREATMENTS

(a) My Auto Advocate Homepage (Opt-in) (b) Competitive E-Brochures

(c) Competitive New-Vehicle Advisor (d) Competitive Community Forum

Page 35: Does Providing Competitive Information to Your …web.mit.edu/hauser/www/Papers/Liberali_Urban_Hasuer...2005/10/25  · (ASIM). ASIM had substantial fixed costs to set up the test

Online Appendix (available from the authors) Comparison of Year-1 Trust Regressions: Treatment Dummies vs. Self-report Dummies

As an illustration, we compare the year-1 trust regressions based on treatment dummies versus

self-report dummies. There are no significant differences (also no significant differences for the age cat-

egories – not shown). This lack of difference is indicative of other analyses with these data in the USAM

experiments and quasi-experiments.

Self-Report Dummies Treatment Dummies Comparison

Effect p value Effect p value t p value

Intercept 0.640 0.333 0.714 0.337 0.156 0.876

Lagged Trust 0.860 0.013 0.857 0.013 0.163 0.870

Competitive Test Drives 0.380 0.081 0.371 0.081 0.079 0.937

Customized Brochures 0.171 0.053 0.127 0.056 0.571 0.568

Competitive Advisor 0.045 0.037 0.016 0.040 0.532 0.595

Competitive Forum -0.057 0.066 -0.056 0.037 0.013 0.989

Own USAM 0.000 0.001 0.000 0.001 0.000 0.999

Own American 0.021 0.038 0.011 0.038 0.186 0.852

Own Japanese -0.019 0.039 -0.023 0.039 0.073 0.942

Period 3 -0.220 0.057 -0.243 0.056 0.288 0.774

Period 4 -0.295 0.067 -0.282 0.071 0.133 0.894

Period 5 -0.127 0.062 -0.119 0.063 0.091 0.928

Period 6 -0.251 0.063 -0.238 0.066 0.142 0.887

Adjusted R2 0.749 0.748