Databases Reference
In-Depth Information
Let's think about a randomized experiment Frank could run. He could
select a bunch of beautiful women, and half the time, randomly, tell
them they're beautiful. He could then see the difference in response
rates between the two groups.
For whatever reason, though, Frank doesn't do this—perhaps he's too
much of a romantic—which leaves us to try to work out whether saying
a woman is beautiful is a good move for Frank. It's on us to get Frank
a date.
If we could, we'd understand the future under two alternative realities:
the reality where he sends out the email telling a given woman she's
beautiful and the reality where he sends an email but doesn't use the
word beautiful. But only one reality is possible. So how can we pro‐
ceed?
Let's write down our causal question explicitly: what is the effect of
Frank telling a woman she's beautiful on him getting a positive
response?
In other words, the “treatment” is Frank's telling a woman she's beau‐
tiful over email, and the “outcome” is a positive response in an email,
or possibly no email at all. An email from Frank that doesn't call the
recipient of the email beautiful would be the control for this study.
There are lots of things we're not doing here that we might
want to try. For example, we're not thinking about Frank's
attributes. Maybe he's a really weird unattractive guy that no
woman would want to date no matter what he says, which
would make this a tough question to solve. Maybe he can't
even spell “beautiful.” Conversely, what if he's gorgeous and/
or famous and it doesn't matter what he says? Also, most
dating sites allow women to contact men just as easily as men
contact women, so it's not clear that our definitions of “treat‐
ed” and “untreated” are well-defined. Some women might
ignore their emails but spontaneously email Frank anyway.
OK Cupid's Attempt
As a first pass at understanding the impact of word choice on response
rates, the online dating site OK Cupid analyzed over 500,000 first
contacts on its site. They looked at keywords and phrases, and how
they affected reply rates, shown in
Figure 11-2
.
