Search the Elsmar Cove!
**Search ALL of** with DuckDuckGo including content not in the forum - Search results with No ads.

Variables Data Type Analysis help


RRMC Steve

A colleague of mine gave me some variable data today that I'm unsure how to assess.

The data are a set of results from a test carried out by 30 different people, using 3 differing techniques, who were asked to give an answer on a scale of 1 to 7.

My colleague is asking if I can tell him if there is a difference between the techniques.

I thought a hypothesis test would be the correct tool but I cannot resolve the issue of the constraints that the 1 to 7 levels might have.

Any help would appreciated.

RRMC Steve

Thanks for responding.

For the sake of clarity I think I need to be a little clearer on the test itself and attached the raw data.

Basically 30 different Appraisers classed into two categories (10 Amateurs, 12 Experts) were asked to give to rate 3 different criterion (Noise, Smoothness & Ease) on 3 different Doors. The ratings they were asked to record were on scale of 1 to 7 (1=good, 7=Bad)

I?ve attached the data as that?s probably the best way to show the results.

Is there a way ? preferably in Minitab ? of looking for significant differences between:
1. Appraisers
2. Criterion
3. Doors

I?ve tried using various Hypo Test in Minitab but without success.

I appreciate any help you or anyone else can give.



Forum Moderator
Staff member
I would start with graphical comparisons such as a Dot Plot by Groups. If this shows something of potential interest, try a rank-based nonparametric test such as Kruskal-Wallis.

There is a lot of disagreement on the validity of analyzing ordinal data as if it were interval/ratio data. In essence, the two faction positions are:
1) Cannot do it because there is only order without evenly spaced scaling
2) Can do it in certain situations where the scale was designed to be as evenly spaced as possible

Many Likert scales fall into the first scenario, but yours may fall into the second. Just realize that a raters perception of a characteristic such as noise cannot duplicate a measurement taken with a meter.

Using a nonparametric test based on ranks will sidestep the problem by using the median. The middle value is the middle value even for an ordinal scale.

There is also a macro available for doing a Kruskal-Wallis multiple comparisons test that provides a nice graphical output.

Bev D

Heretical Statistician
Staff member
Super Moderator
agree with miner. plotting the data to see the comparisons is the best way (we should always do this anyway); despite appraiser measurement differences you should be able to see (literally) if there is any practical difference. if you count, no amount of statistical manipulation will help and if you see a difference statistical manipulation will be irrelevent...

I am away from my laptop or I would plot the data myself.
Top Bottom