The importance of statistics in business (and skinning)

Eek....'assumptions' and 'fudge factors'....of course the stats are gonna be quirky.....you just added the quirk/s....

Factors of 1.1 and 0.9, etc....plucked from thin air, or based on external criteria not related to this specific instance?

Stats can be fun, but the derived results in your blog are not necessarily derived, but 'contrived'.....just as much as the uncertainties of whether the peak popularities were actually externally influenced or not.

It's an amusing diversion but still not a basis for reliable certainty...

Oh, I would say the millions of dollars I've made over the years say differently.

My "fudge factors" weren't arbitrary, as I clearly explained in the article. I know, based on the web logs, which times of years this site is busier than others over a long period of time.

I know that July, for instance, is traditionally a slower month by around 20% than say January. Therefore, modifying the results to de-bias them enables for a better analysis.

I've yet to meet a successful business person who considers statistical analysis a "diversion". I happen to enjoy statistics but it is a major factor in our success.

Facinating article FB...

Ah...but then I wasn't privy to the efficacy of the fudge-factors, nor is/are any of the other blog-readers.

We are left to assume the 'fudging' is an appropriate correction and not merely a book-cooking...

Stats need to be transparent, along with any methods of 'adjustment'.  It is not sufficient to say 'the millions of dollars I've made over the years say differently'....because as the joke goes, you 'may' have made them all 1 inch too long and got caught and are serving time for it....

Hehehe! Keep that up and you might turn into Harry Seldon. When are you going to turn statistics into "psychohistory" and predict the future?
(I'm sure you've read Isaac Azimov's Foundation series)

Busy times need not necessarily relate exactly to WB downloading, as other site sections 'may' adversely influence the site's popularity.

Did any of the sample periods have significant site-access problems?

Was there a public awareness through the general media a factor at one time vs another?

As I said, the 'results' are only as good as the confidence in the accuracy of the math.

'Trust me, I'm almost a Doctor' isn't a basis for confidence in things being depicted accurately....only that the typical issue of 'statistics' again can be shown to portray anything one wishes...

Patric....yes, I've read them.....one of my fave authors...

1. Throwing out #1 and #2 of the top five to get the user base and calling that a "rough method" is quite an understatement.

You should a) use the full pool (the "that would take too long" argument is completely bogus, you toss it into Excel and you're done) and use the right source, for you cannot derive any conclusions about user base from mere downloads alone.

If you'd use your server logs and filter out the downloads in ths section, then you could correlate ip-addresses to downloads and get a far more accurate result.


2. I can see what you're trying to do with that fudget factor, but this is way too vague. You seem to be assuming that a person would download more skins during the dark seasons than in summer.

Jafo, Frankly, I don't really care whether you find statistical modeling or my model accurate or not. The article clearly explains why the summer and winter results were "fudge factored". It even explains why they are to the level they are.

My article is offered for what it is. I am not terribly concerned whether you feel that statistical modeling is something useful or not nor whether you find this particular example to be sufficiently accurate or not.

I am not here to convince anyone of anything. I am merely sharing a tool I have successfully used to help generate millions of dollars in multiple, and very different, markets over the past decade.

Crae: I am "Assuming" that people download more in the Winter because they do download more skins in the Winter. I am applying several years of experience in monitoring monthly downloads of skins overall.  If I was willing to sit down and try to convince you of this, I could also show the bandwidth traffic over a given year.

Similarly, I know that peek traffic hours are between 1pm and 6pm EST. And that Friday is the least busy day of the week despite what seems intuitive on that.

These are all facts that any statistical model would ahve to bias. But what is important, in any profession, is knowing where to draw the line. When the model is sufficiently accurate to serve the purpose it intends.

Perfect is the enemy of good enough. If, in this example, WindowBlinds was losing popularity, it would only be detectable if it was a pretty significant drop.

Using these models, we have been able to track the rise and fall of NeXTStart and ICQ Plus over the years. In our software, our goal is to see whether outside factors are affecting the overall user base and the market as a whole so we can plan for the future.

Fair enough. I can see how monitoring the top 5 and throwing out the top 2 could give you an indication of whether or not an app is on the rise or not.

But still, you do have the full data on all objects, so why not use the full pool. It might also just be that the top 5 gets downloaded disproportionately to the rest of the skins. Who knows.

/me is a perfectionist

On second thought, if it was just meant as an example for use in your article... never mind what I said then.

Crae - exactly. My main point was to put something crudely together to show the importance of statistical modeling in business.

The model I created here was something that could be done very quickly and was still reasonably accurate.

The point in this example was to show how biasing works. Some basic techniques such as zapping the top 2 entries, biasing based on time of years, etc. 

Statistics can be abused in the hands of a marketer but be a very good tool for a business planner.

I've written an article that shows how statistics can be used (and abused) to make business decisions.

Brad...you may not care for my opinion....but you asked for one with the phrase 'and abused'.

Please do not assert that another's opinion is dismissable, particularly when it is actually supporting the concept of 'abuse' by demonstrating ways in which it may have happened EVEN with your own model.

I did not say it was not useful, only that from what was indicated, it has the potential to be flawed....

AND 'something crudely together'....gimme a break....by your own admission it is crude....[unless the preferred interpretation of 'crude' is 'mathematically accurate and statistically fully realised criteria modelled over an extensive period with real-world performance checks']....

Dang it, man....are we not talking about  'the use and abuse of statistics'?

Either your example is a model of abuse, or it is a demonstration of how it SHOULD be done correctly.....

God I hate statistics.....they are tantamount to legalized deception....and why they are so loved by politicians....

The article is about how important statistical models are in having a solid business.

A crude but functional statistical model is still better than no business model.

The corroborating evidence that supports that model, incidentally is that WindowBlinds sales happen to follow that growth pretty closely.

The article also demonstrated some basic techniques for building statistical models - biasing, creating near-means, etc.