Rank confusion
Slow news day


Feed You can follow this conversation by subscribing to the comment feed for this post.


Clearly it was meant to be a frequency distribution, like this, but the writers must have felt they'd had too many bar charts and decided to switch things round a bit.

Rick Wicklin

Although I'm not usually adamant about including zero on the Y axis, when you compare Derek's bar chart with this chart there is a big difference. The fact that the Y axis on this graph starts at 7% instead of 0% badly distorts the vertical scale.


Derek: Thanks for the link. Have you thought about why the profile of the probability is not monotonically increasing as the ranking goes from "very unlikely" to "very likely"?


Kaiser: small numbers. Derek's version shows n=76. The counts on each response only range from 6-12. Flip just a few responses around and the profile looks a lot different. Group them and they smooth out. I'd say there is too fine a response gradation, especially for so few people, and would use a 4-step range.


Thanks for the comments. I finally understood this chart. They plotted the raw data set. They had a 10-point scale, and the numbers are the proportion of responders picking each point. If they labeled the horizontal scale with the points, it would make this clear.

Got confused because most of the time, this type of result is presented as what proportion of responders say "likely" which is interpreted as the top 3 or 4 points on the scale.

I'm also curious that with such a small sample of 67, and with 10 points on the scale, that there would be at least 6 people picking each one of the 10 points.

The comments to this entry are closed.