Internet and the "Long Tail" vs "Superstar effect"

A paraître dans Applied Economics Letters
Internet and the “Long Tail vs. Superstar effect” debate: Evidence
from the French Book Market♦
Running title: Internet and the “Long Tail vs. Superstar effect” debate
Stéphanie PELTIER
GRANEM, University of Angers and LR-MOS, University of La Rochelle, France
François MOREAU1
LIRSA, Conservatoire National des Arts et Métiers, Paris, France
Abstract. From a comprehensive database of monthly sales of comic books and
literature books in France over the period 2003 to 2007, we show that (i) bestsellers
got smaller market shares online than offline, to the contrary of medium and lowsellers; (ii) both online and offline sales shift from the head of the distribution to the
tail with increasing magnitude over the period; (iii) the Long Tail appears to be more
than just a short-lived phenomenon caused by the specific preferences of early
adopters of e-commerce. These three results suggest that online information and
distribution tools, whose use increased over the period 2003 to 2007, do have an
impact on book distribution and on consumers’ purchase decisions.
Key Words: Long Tail, Sales Concentration, Book Industry, Internet.
JEL Classification: D12, L81, L82
♦
This research was supported by a grant from the DEPS / French Ministry of Culture and
Communication and by the French National Research Agency (ANR-08-CORD-018). The authors
thank Gilbert Laffond and Karim Kilani for useful discussions and insightful suggestions.
1
Case 153, 2 rue Conté, 75003 Paris, France [[email protected]]
1
1. INTRODUCTION
Two views exist about the effect of digitization on the concentration of product sales.
Internet might reinforce the popularity of products that are aleady bestsellers
following the superstar effect (Rosen, 1981) or “winner-take-all” phenomenon
(Frank and Cook, 1995). Conversely, thanks to lower distribution costs and new
ways of connecting demand and supply (blogs, forums, recommender tools), a shift
in demand from the most popular products (“hits” or the head of the distribution of
sales) to niche products might occur. This is the Long Tail effect (Anderson, 2004).
Both theoretical works and empirical studies provide conflicting evidence about the
existence and the magnitude of the Long Tail (Bakos, 1997; Brynjolfsson et al.,
2006, 2011; Elberse and Oberzholzer-Gee, 2008; Fleder and Hosanagar, 2009;
Hervas-Drane, 2010; etc.). In this paper, we empirically test the merits of these two
hypotheses using data on the book market. To our knowledge, none of the previous
studies of this industry (Brynjolfsson et al., 2003, 2009; Chevalier and Mayzlin,
2006; Bounie et al., 2010; Benghozi and Benhamou, 2010) is based on such a
comprehensive data set than the one we used, which can simultaneously capture the
evolution of the distribution of sales over a sufficiently long period of time and
compare distributions of online and offline sales, and thus take into account the
increasing use of online distribution and information tools.
2. EMPIRICAL METHODOLOGY
By examining a series of selected quantiles of the distribution of book sales, we can
assess how the sales distribution changes with various covariates that include time,
genre, channel of distribution, etc. It allows us to study especially whether the
possible changes in distribution occur in the head or in the tail. However, a decrease
in units sold for superstars can reflect either a Long Tail effect or a recession
affecting the whole market. To control for this potential bias, we use market shares
rather than units sold as the dependent variable.
The model we estimate is the following:
2
LogQymij = xβ + ε
where Qymij denotes the yth quantile of the distribution of market shares for a given
month m, the distribution channel i (offline or online) and the genre j (literature or
comics). x is a vector of covariates and β is the set of coefficients to be estimated.
The variables used in the regressions are the followings:
-
ONLINE is a dummy that indicates if the distribution of sales refers to online
or offline channels. For a given year, a given month and a given genre, we
indeed consider two distributions of sales (online and offline). Hence when a
title is sold both online and offline it will appear in both distributions.
-
Y2003 to Y2007 are dummies for each of the five years and M1 to M12
dummies for each month.
-
COMICS is a dummy that equals one if the distribution of sales refers to
comic books and zero if it refers to literature books.
-
The log of the number of titles (TITLES) that sold at least one copy in the
month allows us to control for the fact that an increase in the number of
different titles sold mechanically leads to a decrease in the average market
share of all other titles.
-
Since the bulk of sales is made in the first weeks following publication, if the
catalogue of available titles is aging, the head of sales distribution will be
eroded in favour of the tail. NEW is the log of the number of new titles that
have been released during the month and allows us to control for this bias.
To capture the Long Tail effect in the above setting, we make three research
hypotheses.
H1: The distribution of book sales is expected to be less concentrated
online than offline.
The impact of online information tools should indeed be stronger when purchasing
on the Internet rather than in a conventional shop. Furthermore, the use of Internet as
3
a distribution device that makes the access to niche products much easier only affects
the concentration of online sales.
H2: Sales should shift from the head of the distribution to the tail with an
increasing magnitude over the period.
The Long Tail theory assumes that the market share of best sellers should decrease
conversely to low sellers’. The wider use of the Internet as a source of information
and discovery of books is assumed to play a significant role in this change. Since the
use of Internet has increased over the period, the shift in the distribution of sales
should become increasingly important. We thus expect Year dummies coefficients
associated with the highest quantiles to be more and more negative from 2003 to
2007 conversely to lower quantiles. Note that our regressions capture the specific
effect of online sales, and thus the easier distribution effect of Internet. If a
decreasing concentration appears over the period, even when controlling for online
sales and for the number of different titles sold, this should be related to the nature of
Internet as an information device that also affects offline sales.
H3: If the Long Tail effect is not temporary, the lower concentration of
online sales as regards to offline sales should not vanish over the period.
Online purchases could only be due to early adopters who are younger and more
educated than the average reader and who thus exhibit stronger tastes for niche
products (Brynjolfsson et al., 2009). Since the use of Internet has increased over the
period, if the Long Tail effect is not a temporary one, we can expect the lower
concentration of online sales in relation to offline sales not to disappear between
2003 and 2007. We thus include in the previous model Year effects that vary by
distribution channel with five dummies ONLINE×Year. To support hypothesis 3, we
should not observe decreasing coefficients for ONLINE×Year over the period.
4
3. DATA/RESULTS
We use a comprehensive database of monthly sales of physical books (pocket
editions excluded) over a period of five years (January 2003 to December 2007)
obtained from GfK that tracks all book sales in almost all outlets in France. Data
focus on two genres, comic books and literature books which jointly represent about
40% of the whole French book market in 2007. On a monthly basis, the database
includes on average 26,986 different titles of literature books that sold at least one
copy and 11,663 titles of comic books.
The number of monthly observations is 240 (12 months × 5 years × 2 genres × 2
distribution channels), representing 3,200,870 observations at the title level. Each
observation at the monthly level corresponds to a distribution of sales from which we
calculated the quantiles. Table 1 displays the estimation results for regressions on
selected quantiles from the 40th to the 99.9th 2. The effect of a covariate on the
dependent variable varies freely from quantile to quantile.
Model 1 shows a strong and significative impact of the dummy ONLINE on
predicted market shares. Bestsellers (the “top 1%”) perform less well online, though
this effect seems less pronounced for the hits (the 0.1% that sell the most).
Conversely, the market share of low-seller books (belonging to the “bottom 80%”)
are higher online than offline, with the strongest effect for very low-seller books (the
“bottom 40%”). Hence, online sales exhibit a lower head and a thicker tail. Our
results thus support our first hypothesis.
Model 1 also shows that when controlling for the number of titles and for online
sales, the Year dummies still have a significant impact on market shares. For hits, the
decrease in market share is higher over the period and has been significant since
2005. These results support our second hypothesis and suggest that the Long Tail
phenomenon stems, at least partially, from a change in consumer behaviour due to
the use of Internet as an information tool.
2
Regressions on the comprehensive set of quantiles are available upon request.
5
Model 2 shows that for bestsellers (99th quantile), the (negative) difference in online
market share as compared to offline neither increases nor decreases over the period.
For the 80th and 90th quantiles, the magnitude of the positive difference steadily
increases over the period. These results support our third hypothesis asserting that the
impact of Internet on the change in sales distribution is not temporary.
Note that model 2 also gives stronger support to our second hypothesis. The
coefficients of Year dummies (without interactions with ONLINE) reflect the impact
of time for the distribution of offline sales. The signs and significance of these
coefficients confirms that the distribution of sales in conventional shops also shifts
from the head to the tail.
4. CONCLUSION
Our results suggest that the use of online information and distribution tools does have
an impact on consumers’ purchase decisions and leads them to shift somewhat from
bestsellers to medium or low-sellers. Of course, this statistical significance does not
mean that the Long Tail phenomenon in the French book market is economically
very significant yet. In 2007, according to our data, online sales only accounted for
4% of overall sales. However, those sales are experiencing strong growth, which will
be undoubtedly reinforced with the advent of the digital book.
6
REFERENCES
Anderson C., 2004, The Long Tail, Wired Magazine, 12(10), 170–177.
Bakos Y.J., 1997, Reducing Buyer Search Costs: Implications for Electronic Market–
places, Management Science, 43(12): 1676–1692.
Benghozi P.J., Benhamou F., 2010, The Long Tail: Myth or Reality?, International
Journal of Art Management, 12(3): 43–53.
Bounie D., Eang B. and Waelbroeck P., 2010, Marché Internet et réseaux physiques :
comparaison des ventes de livres en France, Revue d’Economie Politique, 120(1) : 141–162.
Brynjolfsson E., Hu Y.J., and Simester D., 2011, Goodbye Pareto Principle, Hello Long
Tail: The Effect of Search Costs on the Concentration of Product Sales, Management
Science, forthcoming.
Brynjolfsson E., Hu Y.J., and Smith. M.D., 2006, From niches to riches: The anatomy of
the long tail, Sloan Management Review, 47(4): 67–71.
Brynjolfsson E., Hu Y.J., and Smith. M.D., 2003, Consumer Surplus in the Digital
Economy: Estimating the Value of Increased Product Variety at Online Booksellers,
Management Science, 49(11): 1580–1596.
Brynjolfsson E., Hu Y.J., and Smith. M.D., 2009, A Longer Tail?: Estimating The Shape
of Amazon’s Sales Distribution Curve in 2008, Working Paper, MIT Sloan.
Chevalier J.A., Mayzlin D., 2006, The Effect of Word of Mouth on Sales: Online Book
Reviews, Journal of Marketing Research, 43: 345–354.
Elberse, A., Oberholzer-Gee F., 2008, Superstars and Underdogs: An Examination of the
Long Tail Phenomenon in Video Sales, Harvard Business School Working Paper No. 07015.
Fleder D., Hosanagar K., 2009, Culture's Next Rise or Fall: The Impact of Recommender
Systems on Sales Diversity, Management Science, 55(5): 697–712.
Frank R., Cook, P., 1995, The Winner-Take-All Society, The Free Press, New York.
Hervas-Drane A., 2010, Word of Mouth and Taste Matching: A Theory of the Long Tail,
mimeo, University of Pompeu-Fabra.
Rosen S., 1981, The Economics of Superstars, American Economic Review, 71(5): 845–
858.
7
Table 1 – Regressions on quantiles
Y2004
Y2005
Y2006
Y2007
COMICS
NEW
TITLES
ONLINE
Log Q40
.147
(.049)**
.232
(.058)**
.450
(.064)**
.655
(.071)**
-.221
(.132)
-.044
(.092)
-2.129
(.092)**
1.156
(.060)**
Log Q80
.024
(.030)
.015
(.036)
.087
(.040)*
.142
(.044)**
.329
(.082)**
-.061
(.057)
-1.300
(.057)**
.329
(.037)**
Model 1
Log Q90
.023
(.025)
.052
(.030)
.111
(.033)**
.132
(.036)**
.294
(.068)**
-.061
(.047)
-1.234
(.047)**
.058
(.031)
Log Q99
-.003
(.017)
.044
(.020)*
.081
(.023)**
.060
(.025)*
-.130
(.047)**
.032
(.032)
-.950
(.033)**
-.286
(.021)**
Log Q99.9
-.074
(.043)
-.117
(.051)*
-.172
(.056)**
-.229
(.062)**
-.200
(.116)
.111
(.080)
-.579
(.081)**
-.270
(.052)**
7.775
(1.036)**
240
3,200,870
0.975
2.375
(.639)**
240
3,200,870
0.977
2.688
(.530)**
240
3,200,870
0.978
2.017
(.366)**
240
3,200,870
0.950
-.089
(.908)
240
3,200,870
0.581
ONLINE×Y2004
ONLINE×Y2005
ONLINE×Y2006
ONLINE×Y2007
Constant
# obs. (month level)
# obs. (title level)
Adjusted R-square
Log Q40
.033
(.062)
.076
(.064)
.385
(.068)**
.685
(.070)**
-.352
(.141)*
-.028
(.087)
-2.363
(.143)**
.754
(.151)**
.322
(.093)**
.483
(.112)**
.343
(.118)**
.200
(.131)
10.060
(1.455)**
240
3,200,870
0.978
Log Q80
.000
(.038)
.000
(.040)
.093
(.042)*
.128
(.043)**
.126
(.087)
-.037
(.054)
-1.665
(.088)**
-.127
(.093)
.194
(.057)**
.299
(.069)**
.319
(.072)**
.430
(.080)**
5.854
(.896)**
240
3,200,870
0.979
Model 2
Log Q90
-.007
(.032)
.027
(.033)
.097
(.035)**
.130
(.036)**
.127
(.072)
-.041
(.045)
-1.533
(.073)**
-.328
(.077)**
.180
(.048)**
.270
(.057)**
.298
(.060)**
.334
(.067)**
5.547
(.743)**
240
3,200,870
0.980
Log Q99 Log Q99.9
.010
-.086
(.023)
(.058)
.038
-.151
(.024)
(.060)*
.077
-.162
(.025)**
(.063)**
.071
-.228
(.026)**
(.065)**
-.130
-.197
(.052)*
(.130)
.032
.110
(.033)
(.081)
-.950
-.574
(.053)**
(.133)**
-.280
-.278
(.056)**
(.140)*
-.027
.022
(.035)
(.087)
.011
.064
(.042)
(.104)
.008
-.024
(.044)
(.109)
-.023
-.006
(.049)
(.121)
2.010
-.129
(.543)**
(1.351)
240
240
3,200,870 3,200,870
0.950
0.576
Notes: * 5% significance; ** 1% significance; Y2003 and ONLINE×Y2003 dummies are omitted; estimations of “Month” fixed effect are
not reported.
8