Hello and thank you for this video. I was just wondering : if the "interest" ordinal variable had more than two modalities (let's say a 7 point likert scale from low to high), should I put it as a covariate instead of a fixed factor ? Thank you
I have conducted an experiment where i evaluate the inhibitory effect of 17 bacteria on 2 phytopathogens, using 2 methods: diffusible compounds method and volatile compounds method, the inhibitory effect was expressed in % of inhibition, each bacterium, was tested 3 times on every phytopathogen (3 replicats) and the inhibition mean was calculated. In this case should i use factorial ANOVA (with 3 independant factors) for the analysis of my data? and what is the most suitable Post Hoc test in this case, Thank you
Thanks for your question. There is no straightforward approach to dealing with a violation of heteroskedasticity when running factorial ANOVA in SPSS. There are "robust" variants of the procedure in other programs (e.g., jamovi allows for robust ANOVA models, which includes the ability to test for significant interaction effects). However, in SPSS this is not available. Some possible approaches to dealing with the problem might be (a) look for potential outliers within your data that might increase the variability in certain cells (btw, in the context of factorial ANOVA, you are testing for heteroskedasticity of the cell variances) and then remove them or transform your data to alleviate the problem (b) test your effects using a more liberal or conservative alpha (note: with more "sharply unequal n's; smaller variances associated with larger groups will increase type 1 error rate; whereas larger variances associated with larger groups increase the risk of type 2 error). Another option might be to model your data through the General Linear Models route (as you would with Anova or factorial ANOVA), go under options, and request Parameter estimates along with parameter estimates with robust standard errors (which will essentially perform the ANOVA parameterized as a multiple regression model with robust standard errors for the regression coefficients). The latter is probably not as intuitive though, presentation-wise. A couple of other things to note: If the ratio of largest cell n to smallest cell n is
Hi everyone, I have a much newer video on this topic with data you can download as well as a Powerpoint presentation at ruclips.net/video/Dg4ho5W4Ygg/видео.html. I hope you all check it out!
Hello and thank you for this video. I was just wondering : if the "interest" ordinal variable had more than two modalities (let's say a 7 point likert scale from low to high), should I put it as a covariate instead of a fixed factor ? Thank you
I have conducted an experiment where i evaluate the inhibitory effect of 17 bacteria on 2 phytopathogens, using 2 methods: diffusible compounds method and volatile compounds method, the inhibitory effect was expressed in % of inhibition, each bacterium, was tested 3 times on every phytopathogen (3 replicats) and the inhibition mean was calculated. In this case should i use factorial ANOVA (with 3 independant factors) for the analysis of my data? and what is the most suitable Post Hoc test in this case, Thank you
Hello, I have a doubt; how will you do it if Levene's test is significant?
Thanks for your question. There is no straightforward approach to dealing with a violation of heteroskedasticity when running factorial ANOVA in SPSS. There are "robust" variants of the procedure in other programs (e.g., jamovi allows for robust ANOVA models, which includes the ability to test for significant interaction effects). However, in SPSS this is not available. Some possible approaches to dealing with the problem might be (a) look for potential outliers within your data that might increase the variability in certain cells (btw, in the context of factorial ANOVA, you are testing for heteroskedasticity of the cell variances) and then remove them or transform your data to alleviate the problem (b) test your effects using a more liberal or conservative alpha (note: with more "sharply unequal n's; smaller variances associated with larger groups will increase type 1 error rate; whereas larger variances associated with larger groups increase the risk of type 2 error). Another option might be to model your data through the General Linear Models route (as you would with Anova or factorial ANOVA), go under options, and request Parameter estimates along with parameter estimates with robust standard errors (which will essentially perform the ANOVA parameterized as a multiple regression model with robust standard errors for the regression coefficients). The latter is probably not as intuitive though, presentation-wise. A couple of other things to note: If the ratio of largest cell n to smallest cell n is
@@mikecrowson2462 THANK YOU VERY MUCH! Incredible response, very useful! I will try with these ideas, THANKS!
@@mikecrowson2462 hello what about breusch pagan test for heteroscedasticity ?? and also why not use a bonferroni correction?
Hi everyone, I have a much newer video on this topic with data you can download as well as a Powerpoint presentation at ruclips.net/video/Dg4ho5W4Ygg/видео.html. I hope you all check it out!