I have some university rankings data where I am asked to analyze whether a higher %of one variable leads to a higher ranking in a particular area. What kind of statistical analysis should I use in excel and how should I report the findings?

  • 2
    $\begingroup$ It is difficult to know where to begin. I wouldn't recommend Excel for any data analysis. If you seriously want help, you need to provide more information: how many universities, what ranking do you have (research, satisfaction...), what variables do you have, is the variable of interest really a proxy for something else that might explain the association (e.g., is socioeconomic status a confounder). I am sure others can think of information they need to begin to help you. $\endgroup$ – doug.numbers Sep 4 '14 at 15:16
  • $\begingroup$ (+1) to above. (1) If this is for fun - excel is great. If this is for a professional project you should use a professional level software (2) software specific requests aren't relevant for this site (3) some minimal effort in phrasing the question and what you've thought of would go a long way $\endgroup$ – charles Sep 4 '14 at 15:25
  • $\begingroup$ You may use SPSS for a regression analysis assuming that (higher percentage) your variable is an independent variable and rankings are your dependent variable. If you have some data for a control variable, you may introduce it as another independent variable and you will get good (here, reliable) results. $\endgroup$ – Subhash C. Davar Sep 4 '14 at 15:29
  • $\begingroup$ This question is too broad to be answerable as stated. You will need to take several (maybe ~3) stats classes &/or work through several textbooks on your own. In the end, the appropriate analysis will be ordinal logistic regression, but this cannot be done in Excel. $\endgroup$ – gung Sep 4 '14 at 16:14
  • $\begingroup$ @gung I am a little surprised at the suggestion to use ordinal logistic regression. In the US, for instance, a university ranking typically is a number varying from 1 up to many hundreds. Wouldn't that therefore require hundreds of parameters (in the form of thresholds for each rank)? Surely more parsimonious models could be found. For instance, if the rankings are based on published scores, then I think you would agree that regressing the scores on any variables used to compute those scores would easily and accurately uncover the way in which those variables were used to compute the scores. $\endgroup$ – whuber Sep 4 '14 at 17:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.