Hierarchical regression question

[edieb]

Advertisement - Members don't see this ad

I was reading a study and it reported that the second block of predictor (a diagnosis of a psychotic disorder) accounted for one percent of the variance in suicide. However, the study reported that psychosis was significant at the .05 level. Is it possible to predict that amount of the DV variance and still be significant???

 

[LM02]

I was reading a study and it reported that the second block of predictor (a diagnosis of a psychotic disorder) accounted for one percent of the variance in suicide. However, the study reported that psychosis was significant at the .05 level. Is it possible to predict that amount of the DV variance and still be significant???

If your sample size is large enough, then yes, you can. Statistical significance does not always equal clinical significance.

 

[Ollie123]

Sure. This is exactly the reason its important to report effect sizes and not just p values (even though a huge portion of the time people don't). If you have a really large epidemiological-style dataset, you can actually have significance with well under 1% variance accounted for.

Though its important to note that things are not "clinically significant" in the sense that it will change how you treat an individual, such things can still have important public health consequences. If you can find some way to alter the variable that causes that 1% variance for half the country's population, it can have a tremendous impact.

 

[cognosco]

I totally agree with the other members regarding effect sizes. The APA publication manual even states that effect sizes should be reported, but they often aren't. Another option is to report confidence intervals, but this is seen even less often, esp. in regression studies. Just my .02.