I meant to do a statistical test on this awhile back, but forgot. Sad, but true. Anyway, I remembered the idea, and thought I'd give it a shot. Unfortunately, I promptly ran into the limits of Excel before recalling that my Regression and Time Series class used specialty software to do its magic. Does anyone out there know of a good, free, and somewhat easy-to-use statistical analysis package?
In consolation to my lack of findings with the seasonal analysis, I did at least find that the Total Hours data fits very well to an exponential curve. However, it doesn't take a statistical analysis to see that - just to quantify it. For those interested, the formula turned out to be
62293.35 + e^(.124155t), where t is months since August 2003
and the significance values were all under 10-40. The saving grace for Intlibber's proposition, however, is the residual plot which seems to demonstrate some autocorrelation that could mean seasonality.
You can find the Google Spreadsheet here. (Scroll down a little to see the graph of the residuals. I'd like to make it prettier, but I'm still learning Google Spreadsheets for now.) Again, anyone who knows of nice, free software for statistical analysis, please comment.
3 comments:
Guardian, I don't know of anything good that is free anymore, unless you can get "minmax" for free (I am really dating myself, as if someone else would! ta dat dat! the drum for the bad joke). A low cost package might be JMP from SAS if they still market it. I use that every now and then for visual analysis, but it also has basic analysis. Your equations are interesting, do you have more data that you could send me? I would love trying out my SAS on it and see what I come up with. Of course if the data is provided by Intlibber, I will automatically add a rather large bias factor! :)
If you have data you are willing to send, please send to arnaud@msgc.biz
Thanks.
Thanks for the ideas! As far as your data request, all of the data I used can be found at the Second Life economic statistics site, here. Just look at the right side and download the key metrics - enjoy!
Hey Guardian,
Try using R, which is free, and install the ts (time series) package which will probably be quite handy. You can difference the series, fit an ARIMA model, look at autocorrelations and partial autocorrelations, look at the spectral density, and predict future values in less than ten lines of code. Give me a shout in-world if you need a hand with it.
- Warren
Post a Comment