How much does a good product rating help a bad product Modeling the role of product quality in the r

1
How much does a good product rating help a bad
product?Modeling the role of product quality in
the relationship between online consumer ratings
and sales

• Wendy Moe
• Associate Professor of Marketing
• R.H. Smith School of Business, University of
  Maryland
• wmoe_at_rhsmith.umd.edu
• Marketing Science Conference
• Ann Arbor, MI
• June 2009

2
(No Transcript)
3
(No Transcript)
4
Online Consumer Ratings

• Potential buyers are increasingly looking to
  consumer ratings and reviews as a signal of
  product quality (Forrester 2008).
• Generally, ratings and reviews reflect product
  quality, and as a result, tend to be correlated
  with product sales.
• Ratings and reviews can often deviate from being
  a good signal of quality
• Strategic manipulation (Dellarocas 2006)
• Social influences (Schlosser 2005)
• Dynamics (Godes and Silva 2006)
• Product life cycle effects (Li and Hitt 2008)

5
Research Objectives

• Examine the role of product quality in the
  relationship between sales and ratings by
  explicitly modeling
• Quality effects on both sales and ratings
• Systematic differences in ratings effects across
  products of varying quality.
• Model the arrival of product ratings which
• Captures social influences and dynamics in the
  ratings environment
• Provides a ratings-based measure of product
  quality

6
Modeling the Effect of Ratings on Sales

• Measuring online word-of-mouth
• Valence, variance and volume
• Inconsistent effects across studies.
• Possible explanations for inconsistencies (1)
  endogeneity and (2) sampling

• Capturing product heterogeneity
• Most studies use cross-sectional data
• None examine heterogeneity in coefficient effects

7
Data

• Sample of 200 rated products was obtained from a
  single retailer of bath and fragrance products.
• Assortment of hedonic and utilitarian products
• Moderately priced items (highest price in sample
  25)
• Private label products
• 48 weeks of data
• Weekly sales for each product
• Continuous ratings and reviews (ratings provided
  on a 5-star scale)
• Ratings tool introduced in the middle of the time
  period (allows us to model baseline sales)

8
Modeling Weekly Product Sales

• Sales is modeled as a function of valence,
  variance and volume, consistent with existing
  literature
• Key differences from existing literature
• Product heterogeneity
• bi Normal(mi, s)
• Effect of relative product quality
• mi m0 gEQi

9
Measuring Product Quality

• Expert ratings and reviews are unavailable for
  most products on the market
• Posted product ratings can provide some
  indication of product quality if we can separate
  the effects of social influences and dynamics
  from that of underlying product quality.
• Model the arrival of positive, neutral and
  negative ratings since the timing and relative
  proportion of positive, neutral and negative
  provides information.

10
Modeling Posted Ratings

• Posted product ratings can be characterized by
  (1) the frequency of arrival and (2) the valence
  of the rating.
• The arrival of ratings is modeled as parallel
  hazard processes for the arrival of positive,
  neutral and negative ratings.
• For each product i, the arrival of the next
  posted rating of valence v is governed by an
  exponential hazard process

11
Ratings Dynamics

• Model dynamics through the use of hazard
  covariates
• Covariates (Xij) LAGVALENCEij
• LAGVARIANCEij
• LAGVOLUMEij
• LNCUMSALESij
• Heterogeneity hazard rate (liv) for each ratings
  level (v) varies across products.

12
Ratings-Based Measure of Relative Product Quality

• Baseline hazard rate (liv) captures the
  underlying propensity to post a positive,
  negative or neutral rating for product i, absent
  of any social influence or dynamics effects that
  may cause a posted rating to deviate from an
  independent and objective evaluation of quality.
• Probability of posting a positive, negative or
  neutral rating
• If we assume that quality ranges from -1 to 1,

13
Comparison Models

• Model 1 Product sales are heterogeneous but do
  not experience differential ratings effects as a
  result of quality
• Model 2 Quadratic effect of quality
• Model 3 Alternative measure of quality using
  simply average rating

14
Benchmark 1 Results
bi Normal(mi, s)
15
Proposed Model Results
16
Effect of Valence by Relative Product Quality
17
EQ as a Measure of Quality

• Correlation with other ratings metrics
• Valence 0.51
• Variance -0.25
• Volume 0.45
• Survey of experienced consumers asked subjects to
  rank order three products in our data based on
  their assessments of quality.
• 1 ranked products mean EQ rank 1.57
• 2 ranked products mean EQ rank 1.79
• 3 ranked products mean EQ rank 2.64

18
Summary and Conclusions

• Two critical issues that must be addressed in
  modeling the effect of ratings on sales
• Heterogeneity in quality across products
• Differential effect of ratings across product
  with different quality
• Implications for empirical findings documented by
  existing literature
• The number of ratings has no significant effect
  on sales
• The effect of valence varies across products of
  varying quality
• Future Research
• Other factors that may influence how a product
  may respond to ratings
• Dynamics in ratings behavior