# Computing Power for ANOVAs using the SPSS GLM procedure

SPSS performs power calculations using analysis of variance on existing data. This facility may be accessed by clicking on the options button in the window which appears after defining a variable in the General Linear Model:Repeated Measures procedure and, for between subject designs, in the General Linear Model univariate procedure. Ticking the box next to **Observed Power** will produce a power computation in the output.

Ticking the box just above called **Estimates of effect size** produces partial $$\eta^text{2}$$ which may also be written as partial eta-squared. When selected these form two extra columns in the outputted anova.

It is informative to note that the power estimates produced in this way by SPSS are more conservative than those suggested, for example, by Cohen. We use Cohen's approach in our power programs. Further details of this discrepancy may be found here which are, also, reproduced in the paragraph below:

"If you look at the SPSS algorithms Cohen's book and do some algebra, you can find that SPSS is obtaining the noncentrality parameter lambda value as DFE*(SSH/SSE), while Cohen is obtaining it as N*(SSH/SSE). The difference presumably comes from the fact that Cohen is treating the whole situation as one in which everything is known, while SPSS is estimating things".