ioi said:
Let me put the ball in your court then. Tell me how you would do it. Here is some hypothetical raw data for three games: Game 1 Week 1 - 984
Week 1 - 477
Game 3 Week 1 - 147 Now assume that the data comes from a sample that represents 0.5% of the total population. Can you demonstrate to me how you would present the weekly chart for those 3 games for those 3 weeks in a way that you feel would make you happy. |
This is really tough to say because of how small the sample is. On one hand, something like game 1 shows a large, clear decline, and so a general estimation showing a decline I think would be fine. Just for easy calculations for show, scaled by 200, for game 1, week 1 published as ~200k, week 2 published as ~80k, and week 3 published as ~40k. The numbers imply a reasonable level of doubt while still giving actual information about the current state of the title. Game 2 on the other hand, the difference between week 2 and week 3 is so small, I have to wonder if there is just some statisical noise involved. For me, I think reporting both as ~70k would be a better bet than trying to discern a trend between the weeks.
I don't think it would be wise to publish the actual starting data, but that could also help show where the published numbers are coming from to also allow the user to make reasonable judgment about the publishings as well.
Like Torillian stated, reporting the 95% CI in addition could also help solve the problem of misrepresentation and show that there is a range of values to consider because of the small starting data.
And of course, the origin of the starting data needs to be explained, as I mentioned before. Not only is it small, but is it actually representative of the whole market? Where and how is this data being gathered?
