Comparing percentages

systat

New Member
I am doing a study, and have gathered the following data:

A patient population was divided into 3 categories:
Young patients (n=19) / Middle aged patients (n=66) / Old patients (n =20)

Based on a questionairre, we have categorized each patient into 2 categories: "Yes family history of diabetes" or "no family history of diabetes".

The data was as follows:
Young patients (n=19):
Yes Family Hx = 7
No Family Hx = 12
So, % of patients with "Yes Family Hx" = 37%

=========
Middle aged patients (n=66):
Yes Family Hx = 21
No Family Hx = 45
So, % of patients with "Yes Family Hx" = 32%

=========
Old patients (n =20):
Yes Family Hx = 1
No Family Hx = 19
So, % of patients with "Yes Family Hx" = 5%

I have used Chi square table to test whether the differences between the 3 groups is statistically significant.

**My question is:
Is there any way I compare these 3 groups using the PRECENTAGE data above? I am asking because I want to represent this data in a graph as follows:
Y axis = % of patients
X axis = Each of the 3 categories of ages as bar graphs.

Most importantly, I want to indicate on this graph whether the % in each age group DIFFER FROM EACH OTHER STATISTICALLY.

A graph similar to what I want to do is found in Figure 1 of this paper: http://circ.ahajournals.org/cgi/content/abstract/74/4/712

medha gokhale

New Member
you can use 'Chi square test for trend in proportions'. you can get detailed procedure for computing in the book entitled"Statistical Methods" by Snedecor G.W. and Cochran W. G. (Oxford & IBH publishing company, 1967). This will help to decide whether the there is an increasing trend in proportions of patirnts as age increases/from low to high age group. become

bukharin

RoboStataRaptor
Your chi-squared test already tells you that the proportions (percentages) differ between the 3 groups. In the figure you cited, all they did was an overall chi-squared test (as you've done) and then repeat the chi-squared test for group 1 vs group 2 and group 2 vs group 3. I wouldn't recommend doing that - I would go with Medha's advice, which is effectively running a linear regression with family history as a binary outcome and age group as the sole covariate; or alternatively run a logistic regression with family history as the binary outcome and age group (or even better, age) as the covariate.