Confidence Intervals for Power and Effect Size for Chi-square Tests

Basic Concepts

The same approach used to calculate a confidence interval for the effect size of a t test (see Confidence Intervals for Power and Effect Size of t Test) can be employed to create a confidence interval for a noncentrality parameter, and in turn Cohen’s effect size and statistical power, for a chi-square goodness of fit or independence test.

Example

Example 1: Find the 95% confidence interval for the effect size w and power of a chi-square test of independence for a 3 × 3 contingency table with sample size 500 when χ2 = 30.

Confidence interval effect size

Figure 1 – Confidence intervals for effect size and power

We see from Figure 1 that the 95% confidence interval for the noncentrality parameter is (9.98, 51.81). The corresponding confidence interval for the effect size w of .24 is (.14, .32) and the confidence interval for power of 99.7% is (71.5%, 99.99%).

Examples Workbook

Click here to download the Excel workbook with the examples described on this webpage.

References

Lowry, R. (2023) The power of the chi-square “goodness of fit” test
http://vassarstats.net/chi_beta.html

Faul, F., Erdfelder, E., Buchner, A., & Lang, A. G. (2009). Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41, 1149-1160.
http://link.springer.com/article/10.3758/BRM.41.4.1149

STAT (2015) Power and sample size reference manual, release 13
http://www.stata.com/manuals13/pss.pdf

Oyeyemi, G. M. (2010) On the estimation of power and sample size in test of independence
https://scialert.net/fulltext/?doi=ajms.2010.139.146#google_vignette

13 thoughts on “Confidence Intervals for Power and Effect Size for Chi-square Tests”

  1. Would the same formula be used for the phi-coefficient? Or is this just for Cramer’s w? I suppose for V the df would always be 2 and this just allows it to be greater than that?

    Reply
    • Venus,
      I have not yet focused on this issue, although I expect to turn my attention to it shortly. I will probably use a simulation approach.
      Charles

      Reply
    • What you have calculated is not the lower and upper limits for lambda, but the lower and upper limits for x. In fact NCHISQ_INV(.025,4,30) = 14.79 and NCHISQ_INV(.975,4,30) = 58.80, the same as the values you have shown.
      Charles

      Reply
      • I cannot understand where is this procedure in Real Statistics. In the list of Data Analysis Tools there is only Staistical Power and Sample Size. I mark it and then mark Chi-Square test and in the menu there are only Size Effect, Sample size, df, alpha and Sum Count (BTW, what is this?). However, there is no option for the CI calculation and no table like your Fig. 1

        Reply
        • Nikita,
          I have not implemented this capability yet in the Real Statistics software. You can calculate it yourself as described on the referenced webpage.
          Sum Count = the number of terms used from the theoretically infinite sum required to calculate the noncentral chi-square distribution value. You don’t really need to worry about this and simply use the default.
          Charles

          Reply
        • Nikita,
          x was simply my way of referencing the noncentral chi-square statistic value. The main thing is that NCHISQ_INV(.025,4,30) = 14.79 and NCHISQ_INV(.975,4,30) = 58.80.
          Charles

          Reply
  2. Dear Dr. Ziontz,
    please, help me to understand is it possible and how to calculate confidence interval for the effect size w using RealStatistics
    Thank you in advance, best regards, Nikita

    Reply

Leave a Comment