# Group comparison: joint model vs. separate model, contradictory result

Any hint on this issue would be really welcome: The goal is to employ a two-country comparison analysis. When running two separate OLS regressions, one for each country, the coefficient of variable x is statistically insignificant in country1 regression (Ho:b1=0 is not rejected), while the coefficient of the same variable x is statistically significant in country2 regression (Ho: b2=0 is rejected).

Afterward, I run a joint model by including a country dummy (country1=1 for country1, and country1=0 for country2) and an interaction term: inter=x*country1. The coefficient of the interaction term appears to be positive and statistically significant, implying that the effect of variable x is greater in country1 than in country2.

So, I am wondering whether there is any statistical explanation for this contradictory result; the separate regression analysis showed that the effect of variable x is insignificant in country 1 and significant in country2, but the joint model analysis showed that the effect of variable x is greater in country 1 than country2...

Thanks a lot!

-
add comment

## 1 Answer

There is no contradiction in what you have written.

The difference between significant and not significant is, itself, not significant.

However, suppose the effect of x in country 1 is exactly 0. That would be as non-significant as you could get. And any effect in country 2 would be different from that in country 1. If the effect in country 2 was big enough, it would be significant.

-
Oh, I understand, thanks a lot for the response. So, it makes sense statistically. It just sounded a bit strange to me, conceptually, that the effect difference between an insignificant (country1) and a significant predictor (country2) appears to be significant in favor of the insignificant predictor. –  Bill718 Aug 3 '12 at 0:34
add comment