Crosstabulation with too many values

[Anne-Claire Jo]

Dear all,


I would like to see the significance of cross tabulation of two variables, however, if I use "tab" command, it says that I have too many values.
Therefore, I would like to ask if there is other method that I can do chi-test or see the significance (option exact for tab command) with too many values?


Thank you,
AS Jo

[Nick Cox]

What are your variables? How many rows; how many columns to your intended table? A contingency table that can't fit under the limits of tabulate is usually one where a chi-square test is of limited value. There could be exceptions, but I would guess that one or other variable isn't categorical if this happens.

[Anne-Claire Jo]

Nick Cox Thank you for your reply !
Actually I would like to do a cross-tabulation of (i) productivity level (non-categorical) and firm size (categorical); (ii) productivity level and digital intensity sector (categorical); (iii) digital intensity sector and firm size etc.
Therefore there are some categorical variables and non-categorical variables.
It's a firm-level panel data so it contains lots of observations. Would it be possible to do a cross-tabulation for the correlation (significance) ?

Thanks

[Nick Cox]

Firm size must be ordered categorical, surely. I don't know what digital intensity sector means precisely. Much depends on whether you are just looking for associations or have an idea on what is outcome and what is predictor.

But (i) and (ii) sound more like regressions with indicator predictors.

Correlations, even rank correlations, don't sound very helpful in any of those.

[Anne-Claire Jo]

Therefore there is no any other solution for using crosstabulation with "too many values" ?

[Nick Cox]

I am struggling to identify the problem here. If one or both variables has hundreds of distinct values, then that implies so many rows and/or columns to a table in principle. What would you gain even if you could produce a table? You could hardly read it and digest it.

None of your relationships strikes me as appropriate for a chi-square test or even for correlation. Regressions seem natural for two of them, in which case a table is irrelevant as graphs will tell you more.

What level are you at? If you are a student, perhaps you need advice from a teacher. If a junior researcher, then seek advice from a supervisor, advisor, or mentor, depending on your situation;

[Anne-Claire Jo]

Nick Cox Thank you first for the advice and your reply.
I thought that it would work with correlation command.
Actually I do not really have someone to ask near me, therefore I asked in this forum.
Anyways thanks for the comment !

[Nick Cox]

Thanks for the thanks, but you haven't answered the question of what you are trying to do.