# 3 columns of categorical data - what test to use for significance?

#### crisisinmedicine

##### New Member
Hi I did surveys on pre-teens, teens, 20s (3 "columns") and need to find signfifiance (p-value). What test do I use? I keep trying Chi square on graphpad prism 5, but I keep getting the message

Analysis of Contingency Table: Overview
Chi-square calculations are only valid when all expected
values are greater than 1.0 and at least 20% of the expected
values are greater than 5. These conditions have not been
met, and thus the chi-square calculations are not valid. To
avoid this problem, combine two or more rows or columns.

What do they mean "combine" I can't just combine three columns of data, so is there a better test I can use? Thank you.

#### trinker

##### ggplot2orBust
Your data is in wide format right now. I'm guessing your program may want it in long format (one column). Here's an example of how the change looks:

Wide format (what you have)
Code:
   preteen teen twenties
1        B    D        A
2        E    B        A
3        D    A        E
4        C    C        E
5        A    A        C
6        E    D        A
7        C    A        C
8        E    D        A
9        B    E        A
10       C    A        B
Long format
Code:
      group out
1   preteen   B
2   preteen   E
3   preteen   D
4   preteen   C
5   preteen   A
6   preteen   E
7   preteen   C
8   preteen   E
9   preteen   B
10  preteen   C
11     teen   D
12     teen   B
13     teen   A
14     teen   C
15     teen   A
16     teen   D
17     teen   A
18     teen   D
19     teen   E
20     teen   A
21 twenties   A
22 twenties   A
23 twenties   E
24 twenties   E
25 twenties   C
26 twenties   A
27 twenties   C
28 twenties   A
29 twenties   A
30 twenties   B

#### crisisinmedicine

##### New Member
Alright I reformatted it that way, now which statistical significance test should I use? For chi-sq it keeps asking for an expected value, and I have none; this was a survey.

#### gianmarco

##### TS Contributor
Hi,
I think that people here could provide sound help as far as you give some more info and elaborate a bit more on your data and the goals you have in mind.

Regards
Gm

#### crisisinmedicine

##### New Member
I sent you a private message because It is a little...private! Thanks gianmarco.