1. ## Survey

I'm a Biostatistician working for a researcher who wants to find out if length of training would affect peoples opinion in choosing a certain medical specialty. He has designed a survey with a 5 point likert scale (Strongly Disagree, Disagree, Neutral, Agree, Strongly Agree) and he ask the question "Would you chose Family Medicine if the length of training is 2 years?", "Would you chose Family Medicine if the length of training is 3 years?", "Would you chose Family Medicine if the length of training is 4 years?". He wants to show that people tend to disagree more with increasing years of training. My questions are; would it be reasonable if I treat the "Neutral" as "Missing" since this isn't adding any information? Can some one suggest a better method for analysis, I have tried the McNemar's Bowker Test for all pairwise comparisons (2yrs vs 3yrs, 2yrs vs 4yrs & 3yrs vs 4yrs) but the conclusions are inconsistent and besides I'm not quite sure if this is gonna answer the question. Lastly, I was thinking of giving each likert a score (1 to 5 for example) such that increasing score indicates increasing agreement then run a repeated measures ANOVA. Any hints

2. ## Re: Survey

Yes, I typically treat likert data as ordinal. You can also collapse the groups disagrees and neutral versus agrees (so binary), you would want to report accordingly and would lose a little of the exactness in these data.

Given that there may be a few options, it all depends on your agenda. I would also say sample size may influence what you do, if there are not many respondents you may be more into collapsing, just report that the effect may be biased toward or away from the null depending on your set-up.

You could also look at a gamma test (correlation of small groupings) or simple ordinal regression might be my recommendation.

3. ## Re: Survey

I typically treat likert data as interval although that makes assumptions that are unknowable. You have to assume that the differences between each level are the same. It does make it much easier to do many analysis and the likert data I deal with (from strongly agree to strongly disagree most commonly) might reasonably have similar intervals IMHO.

The trouble with neutral is that you can interpret this two ways. One is don't know or don't care (or NA) in which case treating it as missing for analysis seems reasonable. Another way to intepret it as a middle value between disagree and agree in which case it's not missing at all - its the third of five values on a continum.

The literature suggests that given odd number of choices like this people chose the middle value often just because they don't want to answer the question. For that reason I use forced choice (only four values) and a don't know NA response.

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts