Hi everyone
This is my first post
I have a statistically significant interaction term on my fixed effects OLS model, for which I've plotted the marginal effects (at values of X1 and X2 +/- 1SD from the mean). X1 and X2 are continuous variables. A reviewer is asking for confidence intervals on the interaction plot. The confidence intervals overlap. This is causing me a bit of a headache. We can interpret the significance of the interaction simply from the p-value on interaction term, given it's OLS. But I need to be able to explain the overlapping confidence intervals.
I looked through some older posts in this form and based on this, I ran pwcompare(effects) and indeed the p-value for the comparison of importance (4 vs 3) is 0.201 (where 4 = High X1, High X2, and 3 = High X1, Low X2). I'm a bit stuck now on how to proceed to offer an accurate explanation of the significance of interaction if the comparison of the marginal effects is not statistically significant. Can anyone provide any guidance? Is pwcompare the correct approach?
In the table: 1 = Low X1, Low X2; 2= Low X1, High X2; 3= High X1, low X2; 4= High X1, High X2
Thanks
John
This is my first post
I have a statistically significant interaction term on my fixed effects OLS model, for which I've plotted the marginal effects (at values of X1 and X2 +/- 1SD from the mean). X1 and X2 are continuous variables. A reviewer is asking for confidence intervals on the interaction plot. The confidence intervals overlap. This is causing me a bit of a headache. We can interpret the significance of the interaction simply from the p-value on interaction term, given it's OLS. But I need to be able to explain the overlapping confidence intervals.
I looked through some older posts in this form and based on this, I ran pwcompare(effects) and indeed the p-value for the comparison of importance (4 vs 3) is 0.201 (where 4 = High X1, High X2, and 3 = High X1, Low X2). I'm a bit stuck now on how to proceed to offer an accurate explanation of the significance of interaction if the comparison of the marginal effects is not statistically significant. Can anyone provide any guidance? Is pwcompare the correct approach?
| Delta-method | Unadjusted | Unadjusted | |||
| Contrast | std. err. | t P>t | [95% conf. interval] | ||
| _at | |||||
| 2 | vs 1 | -.0004031 | .0460109 | -0.01 0.993 | -.0912451 .0904389 |
| 3 | vs 1 | .2478184 | .0549478 | 4.51 0.000 | .1393319 .3563049 |
| 4 | vs 1 | .1816844 | .0565916 | 3.21 0.002 | .0699524 .2934163 |
| 3 | vs 2 | .2482215 | .0789751 | 3.14 0.002 | .0922963 .4041467 |
| 4 | vs 2 | .1820874 | .0452868 | 4.02 0.000 | .092675 .2714999 |
| 4 | vs 3 | -.0661341 | .051558 | -1.28 0.201 | -.1679279 .0356598 |
In the table: 1 = Low X1, Low X2; 2= Low X1, High X2; 3= High X1, low X2; 4= High X1, High X2
Thanks
John
Comment