20th July 2012 at 7:25 pm #2239M. DoogMember
Hello, I have a question about boxplots.
My experiment had 4 conditions. I measured how these groups differed in their consumption on 4 different products. I had to control for hunger (i.e., how hungry people were).
I would like to present the results in a boxplot. Is this possible? Or do I have too many dependent variables?
I wish I could show the means and standard deviations controlled for hunger, but you cannot control anywhere for a covariate in a graph? Or is it possible to create a variable that is corrected for the covariate which can be included in the boxplot?
I hope someone can help me. Thanks in advance!21st July 2012 at 8:30 am #2242Michael CulbertsonMember
It sounds like you want four boxplots (one for each condition) but with values controlled for hunger, instead of your original measurement—do I have that right? To get values controlled for hunger (or anything else), you can set up a linear regression model with hunger as the predictor variable (add in whatever other variables you want to control for, as well). Be sure to use all your subjects—don’t do a separate regression for each treatment condition. Then, have your statistical software spit out the residuals from this regression, and create boxplots based on the residuals, which are the outcome variable, controlled for hunger. Note, the interpretation of the scale is a little different, now. Suppose your outcome variable is how much soup people ate. The interpretation of the original scale is simply how much soup people ate, given which product they had. The interpretation of the new sale is how much more (positive) or less (negative) soup people ate than the average for someone who was as hungry.
Alternatively, if hunger is measured by some categorical scale with only a few values (say 3 or 4), you could simply create one boxplot for each hunger * treatment combination (12 or 16 plots). But, you would only want to do this if it leads to easily interpretable graphics.21st July 2012 at 12:03 pm #2241M. DoogMember
Thanks for the fast reply!
I am using SPSS for the analyses (and graphs). Maybe my question becomes more clear with the attachment:
Above you see my table with results (means of proportional consumption, so a mean of 0.40 = 40% of the total products eaten consisted of product 1). And below a picture of the boxplot that I would like to make (notice that it presents the difference of consumption compared to the baseline group = x-axis).
First I hoped I could make boxplots in Excel solely based on these results. But I would also need the quartiles…
Maybe my data is to complex for such a boxplot?25th July 2012 at 6:19 pm #2240Michael CulbertsonMember
The chart you show here isn’t a boxplot. It’s a bargraph with error bars.
- You must be logged in to reply to this topic.