3 variables in a 2D graph

October 14, 2006 11:45 AM

What's a creative way to graphically visualize user-rating vs time, taking # of ratings into account?

Say the x-axis is months and y-axis is average rating for that month. If one month only has 2 ratings, the plotted point will look as important as in a month where there are 1000 ratings.

Two ideas I've had:

- bigger dots for higher volume months
- plot point every 100 votes, with the x-coordinate of the point at the mean time of the 100 votes (so points won't be evenly spaced)

Can anyone think of anything better?

posted by lpctstr; to media & arts (3 comments total)

- color / intensity of the dots (would have to be more than a point to do this though).

- plot numbers instead of points?

- add a line to the graph for moving average of last 100 votes? It would be fairly easy to see that a point with only a few ratings woudn't push the average at all
posted by true at 11:54 AM on October 14, 2006

How about mean and confidence interval.

You could either use the cumulative confidence interval, so the rating appears to converge on its average over time, or the confidence interval for that month, so that the uncertainty of of each month is clear.
posted by claudius at 11:56 AM on October 14, 2006

The standard way would be to add error bars to your points, using the standard deviation or standard error of the monthly rating. However, in my experience with ratings sites, there's usually not that much spread. My preferred way would be to see all the data, which would mean plotting the time the vote was entered vs. the rating, and maybe drawing a line from month to month representing the average. You'd present something not entirely unlike this graph. You could also just stack the votes on top of one another, not unlike this graph. (Scroll down and hit the make graph button.)
posted by Mr. Gunn at 4:50 PM on October 14, 2006