relative risk vs odds ratio in crosstab

[dave bott]

Hi all,

I'm a dinosaur using SPSS version 8, so let me know if this problem goes
away in a newer version.

I don't usually use 2x2 contingency tables, but multi-variable
regression. Right now I'm trying to compute both odds ratios and
relative risk ratios in a fairly large dataset (7,000 cases).

The help files suggest using

CROSSTABS
/TABLES=var1 var2 BY var3
/FORMAT= AVALUE TABLES
/STATISTIC=RISK
/CELLS=COUNT COLUMN.

The output includes the odds ratios and confidence intervals, but not
the relative risk ratios and/or confidence intervals.

Is there any command that will produce the relative risk ratios and CIs?

I basically took the table data out of SPSS and calculated the RR ratios
in Excel, but I'd much prefer having SPSS compute them (and the CI's).

Any help is appreciated.

Dave

[Richard Ulrich]

On Thu, 1 Jul 2004 14:34:25 -0400, dave bott <spam...@dartmouth.edu>
wrote:

> Hi all,
>
> I'm a dinosaur using SPSS version 8, so let me know if this problem goes
> away in a newer version.
>
> I don't usually use 2x2 contingency tables, but multi-variable
> regression. Right now I'm trying to compute both odds ratios and
> relative risk ratios in a fairly large dataset (7,000 cases).
>
> The help files suggest using
>
> CROSSTABS
> /TABLES=var1 var2 BY var3
> /FORMAT= AVALUE TABLES
> /STATISTIC=RISK
> /CELLS=COUNT COLUMN.
>
> The output includes the odds ratios and confidence intervals, but not
> the relative risk ratios and/or confidence intervals.

Here is most of a Reply I wrote 3 months ago, to a similar question
in sci.stat.edu --

The actual 'RR' is only rarely desirable,
because it is mathematically intractable and
(for most applications) logically indefensible.

If it is going to be large enough so that it differs from the
OR, so that you need it in particular, you will probably
need specialty-software, which starts out by labeling
your groups with total explicitness; instead of something
written for general applications like SPSS.

>
> Is there any command that will produce the relative risk ratios and CIs?

In the interests of good statistical practice: I hope not.

>
> I basically took the table data out of SPSS and calculated the RR ratios
> in Excel, but I'd much prefer having SPSS compute them (and the CI's).
>

Sorry.

--
Rich Ulrich, wpi...@pitt.edu
http://www.pitt.edu/~wpilib/index.html

[Bruce Weaver]

I think this may just be a case of poor labeling of output
by the good folks at SPSS. Here's some output I generated
with version 11:

ZONE * SPRAIN Crosstabulation
Count

|-----------|----------------|-----|
| |SPRAIN |Total|
| |------|---------| |
| |Sprain|No sprain| |
|----|------|------|---------|-----|
|ZONE|Zone A|2 |38 |40 |
| |------|------|---------|-----|
| |Zone B|5 |10 |15 |
|----|------|------|---------|-----|
|Total |7 |48 |55 |
|-----------|------|---------|-----|
¢

Risk Estimate

|---------------|-----|---------------------|
| |Value|95% Confidence |
| | |Interval |
| | |---------------|-----|
| | |Lower |Upper|
|---------------|-----|---------------|-----|
|Odds Ratio for |.105 |.018 |.625 |
|ZONE (Zone A / | | | |
|Zone B) | | | |
|---------------|-----|---------------|-----|
|For cohort |.150 |.033 |.692 |
|SPRAIN = Sprain| | | |
|---------------|-----|---------------|-----|
|For cohort |1.425|.989 |2.052|
|SPRAIN = No | | | |
|sprain | | | |
|---------------|-----|---------------|-----|
|N of Valid |55 | | |
|Cases | | | |
|---------------|-----|---------------|-----|
¢

The first row of the table does indeed give the odds ratio.

OR = (2/38)/(5/10) = 0.105

The 2nd row of the table gives the relative risk of a sprain:

RR of sprain = (2/40)/(5/15) = 0.15

And the 3rd row gives the RR of NO SPRAIN:

RR of no sprain = (38/40)/(10/15) = 1.425

So in version 11, at least, the RR is reported.

--
Bruce Weaver
wea...@mcmaster.ca
www.angelfire.com/wv/bwhomedir/