of The Pirate Bay caused users to increase their visits to other unblocked piracy sites and to cir-
cumvent the blocks through use of VPNs. We found no increase in usage of paid subscription

sites as a result of this block.

5.5 Possible Mechanisms

We motivated our research by describing two primary reasons why blocking access to
multiple piracy sites might have a different effect on consumer behavior than blocking access to
only one site. First, we suggested that the fixed cost involved with switching to a new piracy
sites (which involves both search and learning costs) could be higher and affect more individuals
when more sites are blocked, thus some individuals might choose to substitute legal consumption
for piracy. Second, we considered the possibility of a chilling effect, whereby blocking a large
number of piracy sites sends a stronger signal to pirates about the severity of the antipiracy en-
forcement regime or increases its salience. The question remains whether our results can distin-

guish between these two mechanisms.

Recall that in both 2013 and 2014, we found that the waves of blocks not only decreased
visits to blocked sites but also caused decreases in visits to at least some other unblocked piracy
sites. In 2013 when we had data on the type of piracy sites, we observed that this causal de-
crease was driven by visits to piracy cyberlockers and did not extend to unblocked torrent sites.
If the mechanism driving the effectiveness of blocking multiple sites in 2013 and 2014 were a
chilling effect (based on the broken windows theory of crime and the increased perception of en-
forcement activity), we would expect this to affect illegal behavior at torrent sites and cyberlock-
ers. Because we do not observe a causal decrease in visits to unblocked torrent sites, we infer

that a signaling effect about antipiracy enforcement is less likely to be driving our results.

39

If that is true, then by elimination we are left with the increased search and learning costs
of piracy associated with blocking multiple sites as the mechanism driving our results. This ex-
planation is highly consistent with our results. Some ofthe piracy sites blocked in 2013 and
2014 were popular piracy link sites, which by definition direct users to content hosted on piracy
cyberlockers. Ifthe value ofthese piracy links sites is that they conveniently reduce the search
and learning costs to users for finding content spread across many cyberlockers, then when these
link sites are blocked it could cause a decrease in visits to the unblocked cyberlockers at which
their links pointed. Because link sites operate by pointing to full video files hosted on cyberlock-
ers and not by pointing to torrent sites (which simply index torrent tracker files for P2P down-
loads), we would not expect blocking access to piracy link sites to decrease visits to unblocked
torrent sites. This is exactly the pattern that we observe in 2013, and so we believe the most
likely explanation for why blocking multiple sites has a greater effect than blocking just one
large site is the increase in search and learning costs. In short, we suggest that the drop in piracy
cyberlocker usage in 2013 (and 2014) is the result of making the content on those sites harder to
find by blocking access to a number of convenient link piracy sites, and this story is consistent

with the lack of a drop in usage of torrent sites following the blocks.

3.6 Economic Impact of Website Blocking
While the effect of the 2013 and 2014 waves of blocks on legal channels were statisti-

cally significant, it is important to ask whether they were economically significant. In 2014, we
start with each individual’s observed post-treatment visits to paid legal subscription sites. We
estimate their counterfactual post-treatment visits to these subscription sites by predicting what
they would have been if treatment intensity were zero (our estimate of the counterfactual, or

what they would have been had the individual not been affected by the blocks). We aggregate

40

the difference across all individuals between observed visits to legal sites and counterfactual vis-
its to determine the total causal uplift in visits to legal subscription sites and divide this by the
total counterfactual visits to get the overall percent increase. If we use the coefficient estimate
from the unbalanced panel estimates in Table 5 (0.0104), we find that users of the blocked sites
in 2014 increased their usage of legal subscription sites by 7% relative to what they would have
done in the absence ofthe blocks. If we use the coefficient estimate from the balanced panel es-
timates in Table Al (0.0169), we find that this increase was 12%. As both the balanced and un-
balanced panels have strengths and weaknesses (discussed in Section 5), we suggest that the ef-

fect ofthe 2014 blocks on the usage of legal sites by treated users was somewhere between 7%

and 12%.

Performing the same analysis for the 2013 blocks (but at the group level rather than the
individual), we find that on average the blocks caused treated users to increase their visits to paid
subscription sites by 8% relative to what they would have done if not for the blocks. Thus both

the 2014 wave and 2013 wave appear to have had similar impacts on legal consumption.

It is worth asking if such increases are economically significant, particularly given the
low average visits to subscription streaming sites in our data. We consider our 2014 data and re-
call that a “visit” in our data measures one continuous session at a legal subscription streaming
site, and thus might roughly be equivalent to watching a film or watching one or more episodes
ofa television show. Because treated individuals averaged 2.37 legal subscription visits per
month in the post period, our estimated 7-12% causal increase in visits implies 0.155 to 0.253
more legal subscription visits per person per month. In our sample, 26.3% of users were treated

(used blocked sites at least once before the blocks), and in 2014 the Office for National Statistics

41

reported that there were 22 million Internet connected households”” in Great Britain. If our sam-
ple is representative, then around 5.78 million households were directly affected by the 2014
blocks. Assuming these households responded similarly to our panel, we estimate that the blocks
caused an increase of 896 thousand to 1.46 million legal subscription streaming sessions per
month in Great Britain (and more than that across the UK as a whole). More sessions would lead
to a higher perceived value by users, which would increase their willingness to pay for the ser-
vice, although the precise measurement in terms of demand elasticity is beyond the scope of this
paper.

Of course, it is natural to enquire as to the actual number of new subscriptions caused by
the 2014 blocks. The lowest coefficient on treatment intensity from our logit model on new sub-
scriptions was 0.009 in the unbalanced panel and the highest was .018 in the balanced panel.

The average treatment intensity for treated individuals in this sample was 6.47 monthly blocked
piracy site visits in the pre-period. Holding all other parameters fixed at their mean, we calculate
the logs odds ratio (of becoming a new subscriber) given a treatment intensity of 6.47 versus the
log odds ratio at a treatment intensity of zero. We compute the difference between the two and
convert it to a difference in likelihood of subscribing to legal streaming sites. This corresponds to
a 1.1 (1.5) percentage point average increase in the probability of subscribing for treated individ-
uals in the unbalanced (balanced) panel. In our data, 18.3% of individuals used blocked sites and
did not use legal subscription sites in the pre period. If we assume that this is representative of
the 22 million Internet households in Great Britain, then roughly 4.03 million households in

Great Britain were affected by the blocks but did not have a paid legal subscription prior to the

® Although our PanelTrack data are at the Internet user level, we extrapolate to houscholds for two reasons. First,
PanelTrack generally attempts to capture users from unique households in their sample. Second, decisions on whether
to subscribe to a legal subscription service and how much to pay generally occur at a household level rather than an
individual level, since multiple individuals in a household may use one account.

42

blocks. A 1.1 to 1.5 percentage point increase in probability of subscribing to a service each
month implies an expected 44,000 to 60,000 additional subscribers per month. UK Netflix sub-
scriptions alone grew by 1.9 million between 2014 and 2015°*, and so our implied increase in to-
tal monthly subscribers to all legal streaming services is roughly 2.3-3.1% of Netflix’s growth
that year. We interpret these findings as economically meaningful, given that the monthly price
ofaa subscription to, say, Netflix in the UK is £6.99 to £9.99 per month. Of course, this result
does not account for existing subscription customers who would have otherwise left the service
but were retained as a result of the blocks, or the beneficial price elasticity effects of causing ex-

isting users to view 7-12% more content on these sites.

6. Discussion

While the use of supply side antipiracy actions has increased greatly in recent years as a
tool in the fight against intellectual property theft, there are relatively few studies that have em-
pirically analyzed their effectiveness in changing user behavior. While the studies that consid-
ered the takedown of pirated content found uplifts in legal sales, studies that focused on eutting
off or blocking access to content through a dominant channel found no effect on legitimate con-
sumption. By analyzing the blocking of a single major piracy site in the UK in 2012, we confirm
these prior findings: pirates continued to access illegal content by increasing piracy through other
sites or by finding ways to circumvent the blocks. But we demonstrate that the effect of supply-
side antipiracy policies are more nuanced when more sites are involved: Our results show that
disrupting access to content through a number of the most popular sites causes decreases in over-

all piracy and increases in usage of paid legal channels. And we find evidence suggesting that

 

24 https://www.statista.com/statistics/324092/number-of-netflix-subscribers-uk/

43

the mechanism driving this is that enough sites have to be blocked to sufficiently increase the

search and learning costs associated with additional piracy.

One objection to the causal interpretation of our results might be that legal subscription
sites could have started advertising their services more heavily around the time of the blocks in
2014 and 2013. However, we believe this interpretation is unlikely to explain our results for two
main reasons. First, our difference-in-difference model is able to capture common time trends
through the time fixed effects, so this would only be a concern if legal services could somehow
target high piracy individuals more than low piracy individuals with such advertisements. Sec-
ond, we observe a lack of differential pre-existing time trends, and so this counter explanation is
only relevant if legal services started targeting heavier users ofthe blocked sites (and not lighter
users) with increased advertising, and they did so exactly at the timing of both the 2013 and the
2014 blocks. The discrete jumps in legal consumption following each of these blocks were not
followed by continuing upward trends, and so the timing ofthe correlation between treatment in-
tensity and discrete changes in legal consumption is telling. In short, while no quasi-experi-
mental claims of causality are ever 100% perfect, we suggest that alternative explanations for our

results are unlikely.

There are of course several limitations to this study. First, we were only able to study le-
gal consumption of media through paid legal subscription sites. Users may consume media le-
gally in other ways, such as by digital purchase/rental, physical purchase/rental, or legal free ad-

supported viewing channels. Because PanelTrack observes clickstream data but not actual e-

44

commerce, we cannot infer a la carte purchases or rentals (e.g., people visit a site like Ama-
zon.com for many reasons other than purchasing movies or television).”” Second, because 2% of
ISPs (weighted by market share) did not implement the blocks and the ones that did may have
only fully implemented the blocks by some time in the post period, our results may underesti-
mate the true effect of website blocking on legal consumption. Third, we only observe three
months after each wave ofblocks, and thus we do not know how long our measured impacts
lasted. Although the effects on legal subscription visits appeared persistent in our data, it re-
mains possible that increases in legal consumption caused by the blocks fade over time as con-
sumers eventually identify and grow to trust alternate piracy sites. Finally, we are not able to
fully estimate the social welfare implications of these blocks because we do not know the costs
of these blocks and because we have no data on the long-run impact of increased firm profitabil-
ity on industry output. Future work should focus on these issues to obtain a better understanding

of the broader impacts of site blocking and other anti-piracy measures.

Given the accumulated evidence, how should policymakers view supply-side interven-
tions to curb illegal piracy? We consider by analogy the Greek myth ofthe Hydra, the mythical,
multi-headed beast. The Hydra is one of most difficult animals to kill in Greek mythology. De-
capitating any single one of its heads only results in several more growing back to replace it, an
excellent analogy for our results and those of prior researchers. It is only when a sword is
plunged into its heart that it dies. Removing the source of the pirated content stored in cyberlock-
ers and linked to by many other sites is akin to stabbing the Hydra in the heart (and akin to shut-

ting down Megaupload.com); this is effective but may not always be feasible. Blocking a single

?5 PanelTrack does also track when a user opens the iTunes application on their computer, a common channel for
purchasing digital media. However, many things (including plugging in one’s devices) cause an app like this to open
and so we chose not to purchase these data from PanelTrack.

45

site is akin to decapitating only one ofthe Hydra’s heads. The result will only be a more diffuse
network of piracy sites, with no curb on pirating activity. Blocking multiple sites at once is akin
to decapitating several ofthe Hydra’s heads. With the network of sites significantly disrupted,
this could possibly be a mortal wounding. We have shown that users’ behavior is sufficiently dis-

rupted and that some increase the use of legal channels, and reduce illegal ones.

46

References

Adermon, A.,C.Y. Liang. 2014. Piracy and Music Sales: The Effect of an Anti-Piracy Law.
Journal of Economics Behavior and Organization. 105, pp 90-106.

Aguiar L, Peukert C., and Claussen J. 2018. Catch Me If You Can: Effectiveness and Conse-
quences of Online Copyright Enforcement. Forthcoming, Information Systems Research.

Bai, J and Waldfogel, J. 2012. Information Economics and Policy, Vol 24(3) pp. 187-96.

Bhattacharjee, S., R. Gopal, K. Lertwachara, J. Marsden. 2006. Impact of Legal Threats on
Online Music Sharing Activity: An Analysis of Music Industry Legal Actions. Journal of Law
and Economics, 49(1) 91-114,

Bounie, D., M. Bourreau, P. Waelbroeck. 2006. Piracy and the Demand for Films: Analysis of
Piracy Behavior in French Universities. Review of Economic Research on Copyright Issues, 3(2)
15-27.

Cameron, C., Gelbach, J., and Miller, D. 2008. Bootstrap-Based Improvements for Inference
with Clustered Standard Errors. Review of Economics and Statistics. 90(3), pp. 414-427.

Cameron and Trivedi. Microeconometrics: methods and applications. Cambridge university
press, 2005.

Chen, P., and Hitt, L.M. 2002. Measuring Switching Costs and the Determinants of Customer
Retention in Internet-Enabled Businesses: A Study of the Online Brokerage Industry." Infor-
mation Systems Research, 13 (3), pp. 255-274.

Danaher, B., S. Dhanasobhon, M.D. Smith, R. Telang. 2010. Converting Pirates without Canni-
balizing Purchasers: The Impact of Digital Distribution on Physical Sales and Internet Piracy.
Marketing Science, 2%6) 1138-1151.

Danaher, Brett, Michael D. Smith. 2014. Gone in 60 Seconds: The Impact of the Megaupload
Shutdown on Movie Sales. International Journal of Industrial Organization. 33 1-8.

Danaher, Brett, Michael D. Smith, Rahul Telang, Siwen Chen. 2014a. The Effect of Graduated

Response Anti-Piracy Laws on Music Sales: Evidence from an Event Study in France. Journal of
Industrial Economics. 62(3) 541-553.

Danaher, Brett, Michael D. Smith, Rahul Telang. 2014b. Piracy and Copyright Enforcement
Mechanisms, Lerner and Stern, eds. Innovation Policy and the Economy, Volume 14, Chapter 2
(pp. 31-67), National Bureau of Economic Research, University of Chicago Press, Chicago, Illi-
nois.

Danaher, Brett, Michael D. Smith, and Rahul Telang. 2017. Copyright Enforcement in the Digi-

tal Age: Empirical Evidence and Policy Implications. Communications of the ACM, Vol 60(2),
pp 68-75.

47

Danaher, Brett, Michael D. Smith 2017. Digital Piracy, Film Quality, and Social Welfare.
George Mason University Law Review, 24, pp. 923-938.

De Vany, A.S., W.D. Walls. 2007. Estimating the Effects of Movie Piracy on Box-office Reve-
nue. Review of Industrial Organization 30:291-301.

Dey, Debabrata, Antino Kim, Atanu Lahiri. 2018. Online Piracy and the 'Longer Arm! of En-
forcement. Forthcoming, Management Science.

Donald, S.G. and Lang, K., 2007. Inference with difference-in-differences and other panel data.
The Review of Economics and Statistics, 8%2), pp. 221-233.

Dur, R. and Vollard, B. 2019. Salience of Law Enforcement: A Field Experiment. Journal of
Environmental Economics and Management. Vol 93, Pp. 208-20.

Goldfarb, A. 2006. State dependence at internet portals. Journal of Economics & Management
Strategy. 15(2), 317-352.

Hausman, J., Hall, BH., and Griliches, Z. 1984. “Econometric models for count data with an ap-
plication to the patents-R&D relationship.” Econometrica, 52: 909-938.

Hennig-Thurau, T., V. Henning, H. Sattler. 2007. Consumer File Sharing of Motion Pictures.
Journal of Marketing. 71(October) 1-18.

Herz, B. and Kiljanski, K. 2018. “Movie Piracy and Displaced Sales in Europe: Evidence from
Six Countries” Information Economics and Policy, Vol 43, pp 12-22.

IFPI. 2010. IFPI Response to Commission Green Paper on Creative and Cultural Industries. July
2010.

Kelling, G.L. and Wilson, J.Q., 1982. Broken Windows. Atlantic monthly, 2493), pp.29-38.

Ma, Liye, Alan Montgomery, Michael D. Smith, Param Singh. 2014. The Effect of Pre-Release
Movie Piracy on Box Office Revenue. Information Systems Research. 25(3) 590-603.

McKenzie, Jordi, W. David Walls. 2016. File Sharing and Film Revenues: Estimates of Sales
Displacement at the Box Office. B.E. Journal of Economic Analysis and Policy. 16(1) 25-57.

McKenzie, J. (2017). Graduated Response Policies to Digital Piracy: Do They Increase Box Of-
fice Revenues of Movies?" Information Economics and Policy, 38, 1-36.36.

Peukert C., Claussen J, and Kretschmer, T. 2017. Piracy and Movie Revenues: Evidence from
Megaupload.” International Journal of Industrial Organization 52, 188-215.

Poort, J., Leenheer, J, Ham, JVD, and Dumitru, C. 2014. Baywatch: Two Approaches to Meas-

ure the Effects of Blocking Access to The Pirate Bay. Telecommunications Policy. 38(1) 383-
392.

48