I am using chi-sq to find out the relationship between the nominal variable gender and key variables family environment and vocational identity measured on 5 point likert scale. Should i select the raw scores or recoded values(1, 2, 3 low medium high resp) while running chi square in SPSS.
If you switch from 5 points to 3 points, you'll be losing information, so you would probably want to have a compelling reason. For example, if responses at both of your most extreme Likert scale responses are so rare, you might need to collapse them with the next number so you have enough in each cell. Another possibility is to treat your Likert item like a continuous variable and perform a t-test. While some have sensible concerns about treating Likert scales this way (especially for single items), a chi square is failing to use all your information (i.e., that the responses are ordered). Finally, since Likert scale items are at least ordinal, probably your best analysis would be a Mann–Whitney–Wilcoxon (U) or Kruskal–Wallis test. Best wishes with your study, Samson.. ~ Kevin
Hello Samson,
Since the chi-square test of independence doesn't actually incorporate numeric values (just frequencies within cells), the specific score doesn't matter. However, if you're really asking whether to use all 5 original response categories for the items versus collapsing those into a smaller number of categories (e.g., negative, neutral, positive; or as you suggest, low, medium, high), I think that depends on which representation of the response scale makes most sense to you.
One practical issue is that, for the observed chi-square statistic to track well against the theoretical chi-square distribution, expected cell frequencies (fe) should be 5 or more in each cell. So, the more cells you have (2 levels of gender x 5 response options = 10 cells), the larger the sample size should be to assure that fe is sufficient throughout. In an r x c table, expected cell frequencies are found by multiplying the sum of all frequencies in the given row by the sum of all frequencies in the given column (in which the cell is found), and dividing by the grand total of all frequencies in the table, N.
Good luck with your work.
Dear Samson Ayyamperumal Testing is an analytical tool that is used as the basis for decision making and discussion, both models can be used as long as it is based on the hypothesis that has been formed, so that it can bridge the conclusion process and achieve the objectives of the analysis.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey