1
$\begingroup$

So I had the base of a model that was based on my hypothesis:

Model 1

lm(formula = mean_lookneg_reapp ~ ASRS_Total + 
    AHS_Total + CESD_Total + STAI_Total + AGE + 
    GENDER, data = em_rat_per_final)

In this case, CESD was significant while all other predictors were not.

Then I decided to add another variable in my exploratory analysis:

Model 2

lm(formula = mean_lookneg_reapp ~ ASRS_Total + AHS_Total + 
    CESD_Total +
    **mean_conf_reapp** + STAI_Total + AGE + GENDER, 
    data = em_rat_per_final)

Here CESD was not significant and the new term mean_conf_reapp was very significant. I ran an anova comparing the two and the second was a much better fit of the data. Thats all fair.

However, I then added an interaction between mean_conf_reapp and CESD to see:

Model 3

lm(formula = mean_lookneg_reapp ~ ASRS_Total +
   AHS_Total + **CESD_Total * mean_conf_reapp** + 
   STAI_Total + AGE + GENDER, data = em_rat_per_final)

This model is even a better fit and the interaction term an mean_conf_reapp on its own are significant. I ran an anova comparing model 2 to model 3, and the F statistic was significant. So its a better fit than model 2?

My question is: I will write up that in my hypothesised model CESD was significant and that then as exploratory analysis I decided to add confidence rating and the interaction ... But is this a valid thing to do, or am I accidentally fishing for results? Also, should I then control for multiple comparisons because I created three different models?

$\endgroup$

1 Answer 1

0
$\begingroup$

As long as you explain clearly what you did, then it is OK.

A lot of people are not punctilious about this distinction between hypotheses and exploration. Good for you.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.