1

I have an downloadable app that I promote via various traffic sources. I’m trying to build an “overall quality score” of users I’m getting for each of the traffic sources – this will help me decide which traffic sources I should keep, get rid of, and allocate my spend appropriately. I have provided some sample metrics in the attached 'metrics.png' screenshot.

PartnerID - j-5662

Installs - 6360

Attrition - 65%

IA Clicks - 39159

CTR - 51%

SF Clicks - 4264

PartnerQS - *calculated based on above

Some are percentages, some are integers – in some cases, higher numbers are better, while in other cases lower are better – the end goal is to have some sort of formula that can create a standardized indication of quality. Eventually I’d add some weightings wherein I could rank these metrics – maybe attrition is more important to me than CTR, etc.

I have all my data pulled and now it’s time to start building the formulas. Since these metrics are all different, unrelated values, I’d thought about building another table that would be the quality threshold of each metric Then I could compare the value of the actual metric to the quality threshold – though even as I type this I see the limitations of doing it this way, which is why I’m posting here – just want to get a bit of insight on this.

Thanks in advance

James
  • 11
  • 1
  • Unless you have an outcome variable you want to model, it sounds like a multi-attribute valuation problem: see answers [here](http://stats.stackexchange.com/questions/9358/creating-an-index-of-quality-from-multiple-variables-to-enable-rank-ordering/9361#9361), [here](http://stats.stackexchange.com/questions/9137/quality-price-trade-off/9139#9139) & [here](http://stats.stackexchange.com/questions/3201/how-do-i-order-or-rank-a-set-of-experts/3218#3218). – Scortchi - Reinstate Monica Aug 20 '13 at 08:48

0 Answers0