I have to convert some complicated information at work into a simpler form for the lay person. The information reads something like this (example only):
They wanted to show that tall people eat the same amount as shorter people. When the research was complete, it showed that tall people do eat the same amount as short people. However, when I try and show it in a bar chart, to the lay person, it does look like the tall people eat more. I was not involved in calculating the data and am a good person to work on this because I fall into the lay person role myself. Someone suggested an error bar, but again, even if I have lines showing the confidence values (? right word) how do I explain in lay terms that although the difference in the bars chart make it look like there is a difference (it was a small sample) there is no statistical difference between the two. I need to keep any math or calculations out of it. This is my first time doing this so I want to look good here. Any suggestions? Many thanks!
If you’re using the y-axis to indicate both personal height and calories consumed, then, yeah, it’s going to give that impression.
Partly, it’s because (if both populations take in the same number of calories) then short people do consume more “calories per inch of body height.”
You might benefit from a two-dimensional chart, with body height on the x-axis and calories take in on the y axis. If the populations really are the same, the graph will be level, and that presents the true and correct “impression” graphically.
(You can get as fancy as you want, and either graph a line with error bars, or a distribution of data points, or whatnot. Once you go two-dimensional, basic perceptual errors are harder to fall into.)
So what you’re really saying is you have two factoids:Our tall sample ate 1800 calories on average. Our short sample ate 1750 calories on average. BUT … given 95% confidence intervals, sampling error, and all the rest of the standard statistical mumbo jumbo, it’s really too close to call. The50 calorie difference is not statistically significant.If I’ve correctly understood the situation, you’ve got a hard problem to explain that to simple-minded folks.
If you can keep it in words, the best thing to say is “We ran the experiment and the data is inconclusive.” And stop right there.
The next best thing to do is construct, using proper stats techniques, the limits of your confidence intervals. So it’s really something like “tall people eat somewhere between 1550 and 2050 whereas short people eat somewhere between 1500 and 2000”. Then you can explain that in words as I just did, ending with “… and we can’t be any more specific than that.”
Visually you can do a two-element floating bar graph which shows one vertical shaded bar from 1550 to 2050 and one alongside it from 1500 to 2000. It becomes pretty obvious (at least with my conveniently-chosen made up numbers) that there’s a lot more overlap between the two bars than there is sticking out at the top of the talls or the bottom of the shorts. The point to emphasize is that any spot in either bar is exactly as likely to be the real truth as any other. The middle of the two bars is NOT more favorable or more likely or better than any other spot.
The problem is that is fairly sophisticated statistical thinking. So you may fail to get your message to stick.
Which is why I suggest the best way is not to open the can of worms of trying to visualize counterintuitive data to folks who only have their intuition to interpret the picture you draw. You’re the only one with the sophistication to reduce the data to a conclusion. And the conclusion is “It’s a tie as best we can tell.”