Hi,
I carried out 2 experiments with the same experimental conditions
Exp1 has 3 groups of 8 replicates
Exp2 has 3 groups of 6 replicates
Since I am interested in a dose-effect size, I did an independent sample T-test between each group within the experiment. Normality was OK, equal variances not assumed (between groups).
In Exp1 I can show a significant difference between Group2 and Group3; however in Exp2 this difference is not significant.
How can I combine these two experiments in one SPSS analysis?
I am not interested in the absolute values but rather in the difference in means between groups, since absolute figures are all higher in EXP2 than EXP1.
I don't think I can use ANOVA since equal variances are not assumed? Or does this apply to equal variances within the same group between two experiments?
Thank you for your help!
Procedures, like Welch's, can be applied to ANOVA (e.g., https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/oneway.test). btw, were you adjusting p values?
The ONEWAY command in SPSS also has options to report the Welch and Brown-Forsythe F-tests, neither of which require homogeneity of variance.
- https://www.ibm.com/support/knowledgecenter/en/SSLVMB_26.0.0/statistics_reference_project_ddita/spss/base/syn_oneway_n104fd.html
Another option, though, would be to carry out a 2x3 two-way ANOVA with Experiment as the first factor. If you have a recent enough version of SPSS, you'll find that the UNIANOVA (GLM) command has options for dealing with variance heterogeneity via the ROBUST sub-command.
- https://www.ibm.com/support/knowledgecenter/en/SSLVMB_26.0.0/statistics_reference_project_ddita/spss/base/syn_unianova_robust.html
HTH.
Hello Marlies,
Pooling the data after having run sets of analyses on the separate study batches isn't a great idea. You could treat the two sets of results as independent trials, and compute a collective probability of observing the two specific outcomes (or any pairing you're interested in).
Fisher's method will work here. The formula is simple: -2 * Sum(log(prob)), where log(prob) is the natural log (base e) of the computed p-value for each respective trial comparison you'd like to combine. Treat this as a chi-square statistic with 2 * k df, where k is the number of p-values being combined.
Here's an example (with hypothetical data):
Study 1: p-value = .025; log(p) = -3.6889
Study 2: p-value = .067; log(p) = -2.70306
chi = -2 * sum(log(p)) = 12.784; df = 2*2 = 4
prob(>chi) = .0124
Which implies a low probability of getting two independent results as small or smaller (p-value) than what was observed.
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA