3 way cross tabulation (chi square test)

[drteju]

I am trying to write code in SAS University edition to do a 3-way cross tabulation. For example, if I have 3 categorical variables and I want to look at presisting condition status(preex_status) and insurance coverage(coverage_status) by year whether before or after the policy went into effect (year_class). Each variable has two categories the year_Class has 'pre' and 'post' and the other two have 0 and 1 for yes or no response. I wrote following code:

proc freq data = mydata;

tables preex_status*coverage_status*year_Class / chisq;

run;

However it gives a 2*2 table controlling for preex_status. I cannot use the 'By' statement gives me an error saying "data is not sorted in ascending sequence. I was expecting a table as below:

	Pre	Post
Coverage status	Pre-existing condition	No	Pre-existing condition	No
Yes
No

Is this possible or am I using the wrong statistical test? Can someone help me solve it? Thank yo

[unison]

Looks like a similar question was asked here: https://communities.sas.com/t5/Statistical-Procedures/3-way-cross-tabulations-chi-square-tests/td-p/...

[drteju]

Hi, thank you for the reply. I have tried that solution but does not work in my case. It gives me an error message that, "data is not sorted in ascending sequence'. Thanks.

[unison]

Try running this prior to proc freq:

proc sort data=mydata;
by preex_status coverage_status year_Class;
run;

[drteju]

Thank you @unison that definitely helped and I could run the code. But i still got 2 tables. Is there anyway i could get one table like below:

	Before	Policy	after	policy
Coverage status	Pre-existing condition	No	Pre-existing condition	No
Yes
No

Thank you again 🙂

[unison]

What do you want to go in the blank spots?

if it’s just the frequencies then you can use

out=outfreq

On the table statement (next to your chisq option). Then you would do something like this:

proc report data=outfreq;
column sex height, count weight,count;
define sex/group;
define height/across;
define weight/across;
define count/analysis sum;
run;

[drteju · Accepted solution]

Thanks but that won't work. But i think i might have solved it. I just created a new variable using condition statement and created four categories for the new variable where 1. a person has condition and insurance. 2. Has condition but no insurance 3. No condition but has insurance 4. no condition and no insurance then used this new variable for a two way table with year variable using code:

proc freq data=mythesis.jdata ;

  tables condition_coverage*year_class/ chisq;
  weight perweight1;
run;

and got following output: 

Table of condition_coverage by year_class
condition_coverage	year_class
Frequency Percent Row Pct Col Pct	post	pre	Total
nocond_hascovera	3.364E7 21.26 51.85 42.74	3.124E7 19.74 48.15 39.28	6.488E7 41.01
nocond_nocoverag	5818615 3.68 37.69 7.39	9618986 6.08 62.31 12.10	1.544E7 9.76
preex_nocoverage	3492139 2.21 36.03 4.44	6198915 3.92 63.97 7.80	9691053 6.13
prex_hascoverage	3.575E7 22.60 52.41 45.43	3.246E7 20.52 47.59 40.82	6.821E7 43.11
Total	7.87E7 49.74	7.951E7 50.26	1.582E8 100.00

Looks like that solves the question. Will update once i get it verified if this solution is correct. Thanks again.