Breaking-down main effect to examine demographic differences

I showed videos of musical performances which where visually different, but the sound was identical. Participants graded both the appearance of the performer, and the quality of the musical performance.

I got a statistically significant main effect of visual condition on audience ratings of the musical quality (found the music better under condition #1 than condition #2 even though the sound was identical).

I used a 7 point likert-type scale.

Each participant also completed a short demographic survey including Age, Level of Education, Sex.....

I want to look for whether certain populations were more or less affected by the condition.

But when the data is grouped by demographics, I don't have equal numbers of participants in each sub-pool.

I haven't been able to figure out what test to use.

Can I compare (condition1-condition2) as the dependent variable?

Thank you in advance for any help,
The solution I have tried is for each candidate:

Condition#1 - Condition#2 = Difference

Then I grouped all the Difference data based on Level of Education and did a one-way ANOVA.

For example for each candidate I took the score they gave to "musical performance" under visual condition 1 and subtracted it from "musical performane" under visual condition 2. I did this for each subject then made a table of this item grouped by level of education and ran a one-way ANOVA.

I got a p value around 0.85 but by using the difference between the two scores, the data had a large number of 0, and (+/-)1 making the standard deviation massive.

Here is the data table (note I had not subjects with level of education #2[high school graduates without university degree]):

Groups Count Sum Average Variance
ED LV1 144 34 0.236111111 1.398407148
ED LV2 0 0 #DIV/0! #DIV/0!
ED LV3 72 27 0.375 1.50528169
ED LV4 32 9 0.28125 0.660282258

Source of Variation SS df MS F P-value F crit
Between Groups 0.925963262 3 0.308654421 0.230088615 0.875388944 2.641596472
Within Groups 327.3159722 244 1.341458903

Total 328.2419355 247

The mean condition#1 - condition#2 looks smaller enough for LevelEd#1 that I suspect if I ran more appropriate test, I might get a lower p value (though still probably not significant).

Is there a better test I can do that uses the raw data instead of the condition#1 - condition#2 or is the method I used statistically sound?

Really appreciate any help anyone can provide.