0
$\begingroup$

I have data for an application's installs per day and their ranking in the store for the following day. Each user/installation can be further broken down into two categories - those that installed it because of an ad, and those that installed it on their own (e.g., by word of mouth).

From this data, I want to know if increasing the number of ads would increase installations and thus ranking. I know rank is a terrible statistic to tie it to (since it's not an independent variable; e.g., it's tied to the performance of everyone else's applications in the store), but for the sake of comparison, I have already run a linear regression for the days where the number of ad-driven installs made up for <5% of all installs on a given day.

Given this, which statistical test should I run next to filter out the non-ad-driven installs' effect on the ranking? If I used the wrong metrics (e.g., I should be measuring change in rank against change in installs per day), please let me know.

(And before anyone asks, this is not homework. And on top of that, I know which tests to run after this step.)

$\endgroup$
1
$\begingroup$

I think you've got the wrong data. First, you are assuming that increasing the number of ads would increase the number of installations. If you can get statistics on how many ads were run, seen etc. and how many installations followed that, you might be able to substantiate that assumption. In particular, I suspect that you will find that increasing the advertising for an application that's already advertised heavily does not increase installations as much as for an application that is not advertised. Thus you might conclude for a particular app that there is no point in more ads as the costs would outweigh the benefits.

I don't think that you can separate the effects of ad and non-ad-driven installs (at least not without more data). The two are closely related -- the more ads, the more people will tell their friends about it.

In short, if the main question is "Should I run more ads?", then look at how the number of ads relates to sales. If the question is "Should I focus on ads or promotion through social networks?", compare the numbers for the two kinds of installations.

I agree with your observation that rank is a terrible statistic. I don't think it will give you any valuable insight without further data (e.g. total sales across all apps on a day).

$\endgroup$
  • $\begingroup$ On most days, the number of ad-driven installs is below 25% - I just chose 5% because I would otherwise have only 8 datapoints, and 5% gave a sufficiently large dataset. It's also not heavily advertised. And I, too, have been trying to figure out how to isolate the effects of rankings, which is similar to the chicken-egg problem (i.e., the higher my ranking, the more installs I get, and the more installs I get, the higher my ranking). Though I could run a simulation with the data I have now, I'd feel more comfortable if I could filter out the effect of non-ad-driven installs on ranking. $\endgroup$ – Edwin Feb 3 '12 at 20:48
  • $\begingroup$ I should also add that the advertisers are only paid upon installation, so any served up but not clicked on that resulted in an installation are not tracked. In addition, due to the constraints of the store, it will not forward data indicating what led the user to install the application (only the advertiser gives that info). $\endgroup$ – Edwin Feb 3 '12 at 20:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for?Browse other questions tagged or ask your own question.