# Statistical significance of differences between percentages

I have a problem annoying me. I have associative word lists which were formed by 150 participant's answers to a given word. At the same time, I gathered information about each word's valance with a 5 point Likert type question which ranged from very negative to very positive and at the middle of the range there was neutral choice. Then I grouped these answers into 3 categories (negative, neutral, positive) and I've aggregated the data to percentages. For example, 74% of answers given to valence of words writing for Music list are positive, 16.8% are neutral and 9.3% are negative.

So the question begins here. How can I say these differences between percentages are statistically significant? Which test is appropriate?

• It sounds like the words are in categories of some kind. Is this a variable you're analyzing? You need to give more details of the design with all of the response and predictor variables you wish to assess.
– John
Commented Aug 2, 2015 at 8:02

Use the actual frequency of response in the categories to test for the significance with chi square.

• This seems to be a repeated measures design. Chi-square generally require independent measures. Further, chi-square treats the response categorically, which might be sufficient, but loses out on some power that can be achieved using the ordinal features of the response.
– John
Commented Aug 2, 2015 at 7:58