My variables are categories.

My SPSS result shows that Phi and Cramer's V are positive while Pearson's R and Spearman correlation values are negative.

How do I interpret this?


                   1   2   3
newtypeequity 1    0   0   2
              2    0  35 138
              3    7  18   0

Chi square tests:    Value   df   Asym Sig
Pearson Chi-Square   91.74*   4    0.000
Likelihood Ratio     83.68    4    0.000
Linear by Linear     82.36    1    0.000    

Number of valid       200

* 4 cells have expected count <5, the minimum
   expected count is 0.7

                                              Asymp    Approx    Asymp 
Symmetric Measures:                    Value  Std Err    T       Sig

Nominal by Nominal      Phi            0.677                     0.000
                        Cramer's V     0.479                     0.000
Interval by Interval    Pearson's R   -0.643   0.048   -11.824   0.000
Ordinal by Ordinal      Spearman Corr -0.604   0.050   -10.670   0.000
  • $\begingroup$ Take a look at their scatterplots. $\endgroup$ – Mike Hunter Oct 9 '16 at 20:40
  • 2
    $\begingroup$ Screenshots taken with a phone are essentially unreadable. Please describe your variables (what values do they take? Are they ordered? what do they measure?), how your data were obtained, and explain why you want to look at those correlation measures. If possible extract or construct the data tables and other information as text (indenting 4 spaces) or if you must show images, find out how to get a screen capture into a file $\endgroup$ – Glen_b Oct 9 '16 at 21:37
  • $\begingroup$ I have converted the information that could be read to text, but you didn't show either the top (e.g. the column category variable name) or bottom (the notes under the last table) part of the information. $\endgroup$ – Glen_b Oct 9 '16 at 21:57
  • $\begingroup$ Do you know what a positive correlation means between continuous variables? What a negative correlation means? $\endgroup$ – Glen_b Oct 9 '16 at 21:59
  • $\begingroup$ My variables are level of job satisfaction (low satisfaction, medium satisfaction and high satisfaction) and types of equity ( benevolent, equity sensitive and entitled) . $\endgroup$ – Mee Moore Oct 10 '16 at 6:08

Phi and Cramer's V vary from 0 to 1, whereas the correlations vary from -1 to +1. The correlations are positive when the variables are directly related, (e.g., positive slope of a regression) and are negative when the variables are inversely related (e.g., negative slope of a regression.

Phi and Cramer's V are a lot less commonly used than the correlations. Pearson's correlation, R, when squared, i.e., $R^2$ is the explained fraction, that is, it gives an indication of the strength of the relationship between variables, where $R^2=1$ would be a perfect model relationship, and $R^2=0$ suggests that there is no relationship between the variables.

| cite | improve this answer | |
  • $\begingroup$ Should I use only Phi and Cramer's V value for my research? $\endgroup$ – Mee Moore Oct 9 '16 at 21:09
  • 1
    $\begingroup$ No, see above, however that would depend on what you are doing. $\endgroup$ – Carl Oct 9 '16 at 21:24

The correlations are negative because as one variable increases the other tends to decrease:

jittered scatterplot of categorical data, showing negative relationship

As the variable on the x-axis changes from 1 to 2 to 3, the one on the y-axis changes from 3 to 2-and-3 to 1-and-2 ... clearly as one increases the other is tending to decrease. Hence negative correlation measures.

The nominal measures of association don't have a negative value because they treat the categories as unordered. There's no "direction" to call positive or negative, so values are only more associated or less associated (closer to independent).

You didn't say whether your categories are ordered, so it's not even possible to suggest which might be relevant.

[Now that it's clear that both variables are ordered, neither Phi nor Cramer's V are particularly useful. In spite of the the claim that Pearson's R is for interval-interval association, it could be used here (since the difference between using the original category-scores and the average ranks is not that large), as could the Spearman correlation. I would perhaps lean toward Kendall's tau-b, but it really depends on why you're doing this and what you're hoping to find out.

However, I'd advise you not to even look at a measure if you don't know what it does or why it's relevant (better not to even calculate it if you don't know it's meaningful for your analysis).

It's would be much better to understand what you're trying to achieve and choose your analysis than dump it into some program that blindly calculates dozens of (mostly irrelevant) quantities and then leaves you to pick out what you want. It's direct enticement to data-dredging (conscious or unconscious).

Try to have a goal in mind -- a question or questions you want to address (that is, what you want to find out from your data), and focus your attention toward answering those questions. Exploratory data analysis is fun, no doubt about that, and it can help you to understand your tools, but for it to make sense you need to at least have made some investigation of what those things are and whether they are sensible things to use on your data.

The Wikipedia pages on all of these quantities aren't terrible, so perhaps start with those to get a sense of what they're doing and what they tell you.

| cite | improve this answer | |
  • $\begingroup$ My variables are level of job satisfaction (low satisfaction, medium satisfaction and high satisfaction) and types of equity ( benevolent, equity sensitive and entitled). Postive relation means one variable is increase other one is increase too. My theory is human behavior concern with job satisfaction. $\endgroup$ – Mee Moore Oct 10 '16 at 6:18

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.