Hello all,

I was hoping to get some help from others regarding an analysis I am working on. Briefly, I am conducting a longitudinal analysis on the development of French vocabulary knowledge in a French Immersion setting, with three time points of data. My query is that I have non-significant variance in the slope (which is not too surprising, given the fact that children receive a very similar amount of exposure and instruction in French). My hope is to identify predictors that account for variance in the intercept and slope. Since variance in the slope is non-significant is it possible to accurately identify predictors of the slope? I have talk to others and have heard both sides of the story. Some say, as long as the variance is positive you can still identify predictors, while other say that since it is non-significant I shouldn't even try. Does anyone have advice, suggested readings, or troubleshooted this problem before? Any advice or suggestions would be greatly appreciated.

Cheers!

Adrian

You can use predictors of the slope even though the slope variance is not that different from zero. However, I wouldn't do this and I don't think you'd get much of any information from it. There is no rule against it and it doesn't violate any statistical assumptions, but it is more work and I really doubt you'll obtain any information from it.

For example, see the attached plot. It's a scatterplot of x and y. The variance of x is about 1; the variance of y is about .0001. The correlation is -.02. The regression best fit line is plotted with the scatterplot. You can see there is basically no variability in y, so using x as a predictor of y doesn't really get you anything. It would seem to me a futile endeavor, but it's yours to take if you want to. There's nothing saying you can't.

Thanks for the advice. I really appreciate it.