Dear Sohail,

To test equality of two proportions, the usual test is chi-square.
This is available in SPSS and other widely used software. SPSS like
most packages expects data to be in a rectangular format in which
each line of data represents one individual (n lines for n subjects)
and each column represents one variable. If the data is in this
format, and the two variables of interest are both binary, use
Analyse - Descriptive Statistics - Crosstabs, select the two
variables of interest as the row and column variables, then select
Statistics - Chi-square. SPSS will produce the crosstabulation and
several different chi-square tests - the first one (Pearson chi-
square) is usually the appropriate one. If numbers are small, SPSS
will issue a warning, you should then use the p-value reported for
the Fisher exact test.

If you have a tabulation of the data, and not a file with n records,
set up a new file with 3 columns named 'row', 'col' and 'freq', set
1,1,2,2 in column 'row', 1,2,1,2 in column 'col' and the 4 cell
counts in column 'freq'. Select Data - Weight cases - Weight cases
by 'freq', then get the crosstabulation of 'row' by 'column' as above.

As well as a p-value, you should get an appropriate measure of effect
size, with a confidence interval (CI). The most important measures
used to compare two proportions p1 and p2 are the difference p1-p2,
the relative risk p1/p2 and the odds ratio (p1/(1-p1))/(p2/(1-p2)).
Which is the most relevant to report depends on your research
question and study design. SPSS gives the relative risk and odds
ratio, if you select Statistics - Risk, but you need to make sure
it's calculating it in the right way - there are 8 possible 'relative
risk' ratios that can be calculated for a 2 by 2 table - you need to
swap group labels (1 and 2) for one or both variables, or swap which
you select as the row and which as the column variable, until you get
the right one i.e. the one that agrees with calculating this measure
directly from your p1 and p2.

If you want to report the difference of 2 proportions, SPSS doesn't
calculate this or its CI. I've just uploaded an Excel file that
calculates confidence intervals for proportions and their differences
using recommended methods. You need to use the second block of this
for a difference of two proportions. Simply overwrite the values in
bold with your own figures, the value of p1-p2 and its CI for your
data will then appear in place of those for the specimen data.

In all of the above I've assumed that your data is unpaired. If
subjects in the two groups are individually matched, the appropriate
test is the McNemar test, which is available under Non-parametric
Tests - Two related samples. The third block of the spreadsheet
calculates a CI for the difference. If the odds ratio is more
meaningful, this is calculated as the ratio of the two cell
frequencies representing disagreement - upper right and lower left (b
and c) cells. A CI can be calculated based on one for the simple
proportion b/(b+c), using the first block of the spreadsheet.

Hope this helps.


> > --- In Statisticians_group@..., "sohail"
<sohail@s...>
> > wrote:
> >> I
> > want to
> >> test the hypothesis of equality of two population proportions
> > using
> >> SPSS. I am unable to find the command, someone who knows it, help
> > me
> >> in this regard.
> >> Thanks
> >