# Appropriate method for seeing how similar two items on a survey are?

I have a survey and we want to look at a pair of questions we think are very similar and possibly asking the same thing. If they are we'd like to reduce the two questions to one question. Both questions have ordinal responses ranging from 1 (No) to 4 (Yes, completely). What would be the most appropriate way of seeing if these two items are very similar to where we can reduce it to one item, or if they are not.

For context this is a survey with about 15 total questions, and a total N of ~5000.

Thank you!

• The very first step is to cross-tabulate the answers to the two questions, resulting in a 4 by 4 table. Perhaps you could post that table in your question.
– whuber
Commented Feb 21 at 20:13

If you want to look at difference then you can look at plots of the average difference, the average absolute difference, a Tukey mean difference plot (aka Bland Altman plot) and so on. You can also look at these numerically (except the last). And you can make a crosstab of the two questions and see how often they differ by any given amount (or you could make a table of the differences).

However, you might be interested in the correlation. Then you could use the Spearman rank correlation. This might be closer to what you want, since if the two are highly correlated, they have much of the same information, even if they differ in value. E.g. if everyone scored 1 point higher on one question than the other, there would be perfect correlation even with the difference and you would only need one.

• This was great information, thank you so much!! Commented Feb 22 at 18:23