What expected frequencies should I use?


I'm in the middle of my dissertation and are struggling a little with finding the appropriate expected frequencies in a few Chi-Square tests of single variables.


My hypothesis is that recruiters are moderately to strongly impressed by XYZ.

The null hypothesis is basically that they are not impressed or less than moderately impressed by XYZ

I have data from a Likert scale (7-scale) question, which I chose to be ordinal. 1 signified "Not impressed at all", 4 "Moderately Impressed" and 7 "Very impressed".

Considering the null hypothesis, what would be the best approach for choosing the expected frequencies? Should I just split N among the first three cells (1-3) of the Likert scale and give the rest a frequency of 5 (so that I can actually do the Chi-Square test)?

I don't think that in this case, an equally split frequency among all 7 cells makes sense.

Any help is highly appreciated.



TS Contributor
Considering the null hypothesis,
You have not yet formulated a null hypothesis. A null hypothesis (the hypothesis
to be nullified) is a precise statement, e.g. "the median level of impressiveness
is < 4 [in the population]", or "> 30% of recruiters are less than moderatly
impressed". You can then use an appropriate one-sample test (maybe Wilcoxon
signed rank test in the first example, one-sample Chi² in the second).

By the way, for Chi² not the ACTUAL cell frequencies, but the EXPECTED cell
frequencies have to be >= 5.

With kind regards