0
$\begingroup$

For my thesis i have to find out if there is any statistical correlation between a Traditional+Experimental teaching method (compared to a merely traditional one) and a highier level of understanding of certain subjects.

I used a survey structured on a likert scale of 4 (Strongly agree/disagree), with two sets of questions for each subject.

  • TRAD. lessons completed
  • First survey: "TRADITIONAL teaching method has helped in the
    understanding of x" "i strongly agree/disagree"
  • EXP. lessons completed
  • Second survey: "EXPERIMENTAL teaching method (on top of a traditional one) has helped in the understanding of x"

Where x are 4 different kinds of specific notion/subject.

I used fisher test because i have a low total count(8 people). Code in R, with the first set of data for x1, looked something like this

input_mxn_table = structure(list(Standard = c(0L, 5L, 2L, 1L), Tread = c(1L, 5L, 1L, 1L)), .Names = c("Standard", "St+Experimental"), row.names = c(NA, -4L), class = "data.frame")

fisher.test(input_mxn_table)

The results look realistic as in 3 out of 4 subjects there is no improvement that is statistically relevant. But does this make sense? Did i make any mistake (logic or code)?

Thank you for your time.

$\endgroup$
4
  • $\begingroup$ And what are the rows (1, 2, 3, 4)? Are those the four subjects? Or are those the four Likert response categories? $\endgroup$ – Sal Mangiafico Jul 25 '18 at 17:39
  • $\begingroup$ The same eight people took both surveys? $\endgroup$ – Sal Mangiafico Jul 25 '18 at 17:40
  • $\begingroup$ The table is a table of counts, not Likert scores? $\endgroup$ – Sal Mangiafico Jul 25 '18 at 17:46
  • $\begingroup$ 1) the four response categories (strongly agree/disagree) 2)yes 3)Counts, how many people answered with one of each of the four responses(for each of the total subjects) Sorry if i wasn't clear enough $\endgroup$ – Bos Jul 25 '18 at 21:06
3
$\begingroup$

One challenge with your analysis is that the same eight people took both surveys. It would be better if your analysis reflected the paired nature of the observations --- both for theoretical reasons, and probably for practical reasons.

Another consideration is your treating the Likert response categories as nominal, when they are really ordinal. It is often advantageous to treat ordinal data as ordinal so that the information on the ordering isn't lost.

Given those considerations, it might be better to use a paired sign test or a Wilcoxon signed-rank test. The latter assumes interval data, so the spacing between the Likert categories would have to be specified.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.