Interpretation of p-value generated by the command "table1"

[eric chen]

Hi,

I found this command "table1" on statalist, which allows user to generate a table of descriptive statistics. It also reports a p-value when observations are grouped. For example, when the variables I included in are categorical, the p-value is generated using Pearson's chi-squared to compare characteristics between groups.

table1, by(gender) vars(affair cat \ HAPMAR cat)

In this case, the comparison is between male and female. It reports the following table:

How do I interpret the p-value here?

Thanks

[Joe Canner]

Eric,

The p-values from the table1 command are most likely the chi-square p-values from an underlying tabulate command with the chi2 option (for example: tab gender affair, row chi). The chi-square test is a measure of association which indicates whether the frequencies in a cross-tabulation are more different than one would expect from the marginal frequencies of each variable. In this case, the test shows that gender 1 is more likely to have an affair (13.8% vs. 7.0%) and that gender 1 is less likely to be "IAP" and "Not too happy" but more likely to be "Very Happy".

Regards,
Joe

[eric chen]

Eric,

The p-values from the table1 command are most likely the chi-square p-values from an underlying tabulate command with the chi2 option (for example: tab gender affair, row chi). The chi-square test is a measure of association which indicates whether the frequencies in a cross-tabulation are more different than one would expect from the marginal frequencies of each variable. In this case, the test shows that gender 1 is more likely to have an affair (13.8% vs. 7.0%) and that gender 1 is less likely to be "IAP" and "Not too happy" but more likely to be "Very Happy".

Regards,
Joe

Hi Joe,

Thanks for the reply. So if the p-value is big, say 0.5, does that mean the corresponding variable is not statistically different between two groups (i.e. gender 1 v.s. gender 0) ?

[Carlo Lazzaro]

Eric:
if the p-value is higher than 0.05 (the customary cut-off), there's no evidence of gender difference for the variabe under investigation (obviously, there may be a real diference between men and women but your sample is too small to detect it: see, if interested, my favourite reading on this matter:http://www.bmj.com/content/311/7003/485)
Taking the matter further, a quite famous paper on the statistics concerning the topic you're researching about can be found at http://people.stern.nyu.edu/wgreene/...talAffairs.pdf