1

Possible Duplicate:
What are good basic statistics to use for ordinal data?

I have gathered data from about 70 construction firms who are not currently using a specific software and I asked them why they have not implemented such a software based on 5 point Likert scale. The variables were, high cost, long time required for implementation, difficulty of using the software, etc. For example they have to choose whether high cost of the software was the reason that they have not installed it by selecting strongly disagree, disagree, neutral, agree, strongly agree.

My question is that how I can rank those 5 main variables (cost, time, etc.)? Can I just provide the mean score and rank them in terms of the mean for answers?

Community
  • 1
Ali
  • 31
  • 1
  • 2
  • This kind of question is fairly common & has been addressed a couple of times here on CV. For the best treatments of the underlying issues, you should read the following questions & the answers by Jeremy Anglim: [what-are-good-basic-statistics-to-use-for-ordinal-data](http://stats.stackexchange.com/questions/97/8995#8995) & [is-amazons-average-rating-misleading](http://stats.stackexchange.com/questions/31598/). If you still have any questions left over after having read these, edit your Q w/ what you still need to know. – gung - Reinstate Monica Sep 02 '12 at 14:50

1 Answers1

2

You can rank them in many ways.

Whether you can validly take the mean is an interesting question. Taking the mean implies that the gap between each level on the Likert scale is the same, in some sense, as the other gaps. That is, the difference between (say) "strongly disagree" and "disagree" is the same as that between "disagree" and "neutral". Taking the median makes no such assumption, but with only 5 levels, the medians may not vary much. You could also rank them by something like "% of people saying 5".

Given that you have only 70 firms, you may want to look at each one in more detail.

You may also want to see if the reasons cluster somehow - do people who rate one variable highly also rate another one highly?

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • Yes, there is always a dispute whether to consider Likert scale ordinal or interval. I want to know is there any specific parametric or non parametric test to rank those variables based on the responses provided? – Ali Sep 02 '12 at 12:51
  • [This article](http://dl.acm.org/citation.cfm?id=1241193) may be helpful. – Peter Flom Sep 02 '12 at 13:10
  • "You may also want to see if the reasons cluster somehow - do people who rate one variable highly also rate another one highly?" Can I use Spearman's correlation test to find this kind of relationship? – Ali Sep 03 '12 at 00:24
  • No, you would have to use cluster analysis, if you wanted to do it formally. – Peter Flom Sep 03 '12 at 01:02