IV going from p > .05 to p< .05 in Hierarchical regression?!

#1
Hi all,

So I was running a Hierarchical multiple regression with 4 IVs in the first block, then 3 IVs in second block. the Model for the first block showed all IVs as statistically insignificant. But then, in the second block, one of the IVs then became stat sig with the introduction of the other 3 IVs (only 2 of which were stat sig themselves).

So I ran an moderation analysis and mediation analysis. No moderation. But mediation showed that one of the block 2 IVs mediated the original block 1 IVs effect on DV. But the mediation analyses suggested this was only partial mediation - the data showed a c prime route for effect on DV.

So my question is - HOW? How can the mediation analysis suggest a c prime pathway, when the HMR only had block 1 IV become stat sig AFTER introduction on block 2 IVs?

I hope someone can help.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Do you have any theory to support the models our is this just an exploration excursion. If the latter, you will find all kinds of things, but it doesn't mean they are true and will generalize.

Giant corn - like Peruvian varieties that I can get in corn nut form and in my ceviche?
 

Karabiner

TS Contributor
#3
Do you have any theory to support the models our is this just an exploration excursion. If the latter, you will find all kinds of things, but it doesn't mean they are true and will generalize.
Agreed. The original post seems to include at least 3 distinct models, one hierarchical analysis, and dozens
of significance tests. It looks quite futile to make sense out of the pattern of results, on the basis of "this here
is significant, that there not significant" (besides, no sample size or any actual regression weight reported).
 
#4
Do you have any theory to support the models our is this just an exploration excursion. If the latter, you will find all kinds of things, but it doesn't mean they are true and will generalize.

Giant corn - like Peruvian varieties that I can get in corn nut form and in my ceviche?
Haha giantcorn the beer snack (Japan). Yes block entry of variables was based on theory (from other literature), i placed controls in block 1 - so model 1 was controls. Model 2 included experimental IVs, which is when one of the block 1 controls became statistically significant.
 
#5
Agreed. The original post seems to include at least 3 distinct models, one hierarchical analysis, and dozens
of significance tests. It looks quite futile to make sense out of the pattern of results, on the basis of "this here
is significant, that there not significant" (besides, no sample size or any actual regression weight reported).
hello Karabiner. thanks for your input. Sample size was 86. Calculations indicated that at p .05, anticipated effects size 0.15-0.20 I could just about support 6 IVs. So i had 4 controls and 3 experimental IVs. I just used the basic SPSS linear regression function so I assume unweighted regression?

However you are right that, clearly more research is needed but that something *may* be happening.