I am trying to analyse data collected through the Self-Assessment Manikin (SAM), which uses three pictorial scales to assess the three dimensions of pleasure, arousal and dominance. Participants are explicitly instructed that they can tick either of the five pictures or any of the four spaces in between, making it essentially a 9pt Likert item.
Unsurprisingly, however, people mostly tick the boxes themselves. Out of 96 responses, only 5 ticked any of the spaces in between. I am wondering if this is a problem that can influence my results. So far, I've tried collapsing the data to a 5pt scale by aggregating
9. Is this a sensible step to take before applying any of the usual Likert statistics and visualizations or is it better to leave the data as is? Does collapsing the scale to 5 values give me a better justification to treat the data as interval?
function(x) ifelse(x > 5, floor(x/2)+1, ceiling(x/2))
In general terms I'd suggest it's better to leave the data `as is' for data analysis (so keep the even-number values in), depending on what you want to do. 5 out of 96 responses with inter-box responses makes it unlikely that it will have much influence on your measures of central tendency or variability anyway
(You can check this yourself by comparing the summary statistics for the original data to the same statistics when rounded to odd-values only.)
And on a last point, I don't think that collapsing to 5 response options gives any better justification for treating the data as interval — if anything, I'd think that having fewer response options makes for less-usable as interval data. See chl's response to a question on when one might treat ordinal data as though they were interval.
I'll just add that I used these SAMM scales many years back in my first year of my PhD — honestly can't remember what I did with the between-the-boxes issue (it was a computer based interface, so you might not have been able to select the spaces anyway…)
So +1 to the question from me, for the fond memories as well as the points raised.
- Solved – How to rescale two questionnaires that use slightly different response scales in order to make them comparable
- Solved – Two different levels of Likert scale
- Solved – Is likert type scale interval or ordinal?
- Solved – Which statistical test should I use – likert scales and multiple categorical groups
- Solved – Reporting the use of different likert-scales in survey for structural equation modelling