Calculating post-hoc power

[Hannah]

Dear all, 

I am running a pathway analysis in Lavaan and am trying to compute post-hoc power using G*Power 3. Here is the model I have run and part of the output:

model2 <- '

# regressions

AASP_SAvo ~ SRS_T + BPQ_T

AASP_SSens ~ SRS_T + BPQ_T

BPQ_T ~ SRS_T

TAS_T ~ SRS_T + AASP_SAvo + AASP_SSens + BPQ_T

IUS_comp ~ SRS_T + TAS_T

CERQ_Negative ~ TAS_T

LSAS_T ~ SRS_T + AASP_SAvo + AASP_SSens + TAS_T + CERQ_Negative + IUS_comp

# correlations

AASP_SAvo ~~ AASP_SSens

'

fit2 <- sem(model2, data = Dat_final.cep, estimator = "MLR", missing = "fiml", fixed.x = F)

summary(fit2, fit.measures = T, standardized=TRUE, modindices = T)

To compute post-hoc power using G*power 3 (in χ² tests - Generic χ² test) you need the following information 1) Non-centrality parameter 2) a err prob and 3) Df. The a err prob = .05, the df from my model above is 9, however, I am unsure which statistic refers to the non-centrality parameter?

I know that the non-centrality parameter = normal theory weighted least squares χ² - df of your model. Does this mean that I have to run a model with the WLS estimator to get the normal theory weighted least squares χ²  to calculate power in G*power or is there something I am missing from my output above? 

Also any alternative methods for calculating post-hoc power are welcome!

Thanks!

Hannah

[Terrence Jorgensen]

I am unsure which statistic refers to the non-centrality parameter?

This isn't a software question, so SEMNET is more appropriate.

http://www2.gsu.edu/~mkteer/semnet.html

X2 - df is generally the estimate of the NCP, but post hoc power analysis using your observed sample stats is pretty useless for telling you anything more than whether your p value was less than your alpha level.  Power analysis is more meaningful when you base the effect size (NCP) on a meaningful criterion, such as specifying as a set of population parameters that differs from your null hypothesis to the minimal meaningful degree.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[Hannah]

Thanks Terrence. That's really helpful!