alternative to ANOVA?

#1
Hello, for a consumer behaviour experiment I had one predictor variable (control and treatment groups).
The "treatment" consisted in mortality reminders, and the effect to be measured the impact of mortality reminders on brand preferences.
The participants in the treatment condition were furthermore asked questions about the psychologic impact of those mortality reminders.
I found that the effect of the treatment increased for the treatment respondents who had rated mortality reminders more disturbing (on a Likert scale). However, I cannot use this variable as a moderator in linear regression since it is not included in the control condition.
A solution could be dividing the treatment group into two groups (high and low mortality disturbance).
Then I could proceed with ANOVA (3 groups: control, high and low mortality disturbance).
Are there better solutions?
Thank you for your help.
 
#2
I found that the effect of the treatment increased for the treatment respondents who had rated mortality reminders more disturbing
So you could report this as an additional finding, without adjusting the preplanned
analysis to it? Your p values would be wrong anyway, if you eyeballed results for promising
predictors, and included them into the model as if no such eyeballing was performed.

With kind regards

Karabiner