bottom ranks

[Rajiv Garg]

 We understand that that the model was built based on the TOP 200 apps selling on app store. We did a distribution fitting using @Risk according to the formula you derived. The model fits a LogNormal distribution perfectly based on the Anderson-Darling statistic, which is what it is supposed to be, as your model is based on the Pareto distribution. However, this only holds for up to 200 data sets, and when we extend to more than 3000, the model would not fit. As you might know, there are more than 130,000 game apps on the apple store, we wonder to what extent do you think your model can predict the number of downloads according to the ranking? 

 
We endeavoured hard to exploit the publicly available data, but as a matter of fact, they are very limited. By the time we are fitting our distribution for the Android apps, we cannot move forward. As we discovered that you have estimated the Pareto parameter on Android system but no scaling parameter is estimated. We wonder if any information can be provided to us for further investigation (e.g, scaling parameter, or the number of total downloads for each app)?

 

 

DY, lse.ac.uk

(4/10/13)

[Rajiv Garg]

 I am glad you found our work useful. 

As you guessed the model works only for top 200 apps at this time because those are published by Apple. Similarly for Android you have 480 ranks published and you can infer demand to that number. Beyond those published lists we believe that the demand is less dependent on rank but more on keywords and external publicity thus the model won't fit well to unpublished ranks. You can possibly estimate by tracking the ranks within categories, but finding the top grossing rank by category might be hard. 

To estimate the scale parameter for Android we needed the aggregate sales of apps during some time period, which was unavailable. Thus, we couldn’t estimate it quickly in that paper. There are 2 ways you can use to estimate the scale parameter for android:

1. Find and contact a top app developer to share one (or many) day sale data with you

2. Look at any popular app that released recently and made in the top ranks. Follow that app for a few days and track the aggregate download numbers and ranks. Use those to estimate the scale parameter. You want to track because the range of download will give you some error but over multiple days, the error will be reduced.

 

RG

[Rajiv Garg]

Not many researches have been done in this realm since there is very limited available information. I wonder if the factors  "keywords and external publicity"  are found through your study, or are they inferred from your research.  I am not sure how many data you are possessing, but intuitively, I think the number of downloads/demand should be influenced by the marketing efforts(reflected by promotion spending), and the quality of the app(reflected by the customers rating). These two might create positive correlation but the effect might decay over time. We do not have sufficient data to justify this. Nevertheless, we want to ask if you have a sales curve for a particular app or for several apps for a certain time period, and I think that might be useful for us to conduct further statistical analysis.

 

DY, lse.ac.uk

(4/11/13)

[Rajiv Garg]

Yes, marketing efforts and app quality will influence the demand. I am finishing a work that presents the effect of app quality on demand and should be ready by end of May. I don’t have a paper currently available on externalities/marketing on app store. That project is in very early stages. I will share when I have something ready.

 

RG