5
$\begingroup$

I wonder whether any of you can help me. SPSS puts confidence intervals in graphs of frequencies and proportions, but it clearly is not using a normal approximation, i.e. it is not using the formula $\text{CI} = m \pm 1.96 \sqrt{(p(1-p)/n}$. apart from anything else the SPSS interval is not symmetrical around the value.

So what is it doing? I could not track down the formula anywhere, not even searching on the web (which usually solves 99% of problems in life, plus or minus 1%). :)

Improve this question
$\endgroup$
2
  • $\begingroup$ Can you provide some detail on what graphing function you are using (menu options etc.) for getting these graphs? I was very much under the impression that SPSS doesn't calculate confidence intervals for proportions... $\endgroup$
    – James Stanley
    Commented Nov 19, 2012 at 21:31
  • 1
    $\begingroup$ Also a note on confidence intervals on proportions -- most of the preferred methods for calculation will return an asymmetrical confidence interval (e.g. using the binomial distribution). See the OpenEpi website under the "proportions" option for some different calculation methods for CI for proportions, which you could cross-check with your SPSS results. $\endgroup$
    – James Stanley
    Commented Nov 19, 2012 at 21:33

2 Answers 2

Reset to default
3
$\begingroup$

Statistics graphics use Jeffreys binomial for the CIs.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Well that's interesting to see CI in the graphing dialogue -- can SPSS report numerical values for these CIs for proportions in a frequency table? (Excluding the possibility of using any add-in packages for reasons of convenience/practicality for students.) I keep telling students that SPSS doesn't calculate confidence intervals for proportions and would be slightly happier to be proven wrong. $\endgroup$
    – James Stanley
    Commented Nov 20, 2012 at 2:27
1
$\begingroup$

If you paste the syntax from the Chart Builderfor an error bar chart, you can add labeling syntax to it. For example, the error bars as pasted from the CB would be specified like this: ELEMENT: interval(position(region.spread.range(jobcat*(LOW+HIGH))), shape.interior(shape.ibeam))

Adding the label function as below will cause the values to be displayed. You might need to increase the number of decimals displayed by using the CE. That could be saved as a chart template.

ELEMENT: interval(position(region.spread.range(jobcat*(LOW+HIGH))), shape.interior(shape.ibeam),label(LOW),label(HIGH))

There is also an extension command, PROPOR, that will do all this in a tabular way. However, it requires that the Python Essentials and the PROPOR command be installed from the SPSS Community website (www.ibm.com/developerworks/spssdevcentral)

You can get bootstrapped ci's using Bootstrap and Frequencies. See the bootstrapping case study in help for details.

And you can get a Wald CI from GENLIN using an intercept-only model.

HTH, Jon Peck

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.