LSD and DUNCAN test

[Zeick]

Hello everyone, I have discovered this software and I am pleasantly surprised. It is very easy to use, with a very simple and beautiful interface.

I have a series of data in different categories that I would like to analyze with ANOVA to see significant differences. However, I am looking for the possibility of carrying out post-hoc tests and I only see the possibility of carrying out the Tukey test.

I would like to know if there is any possibility of carrying out the LSD or DUCAN test, since in some experimental designs they are the ones that best fit.

On the other hand, when presenting the results, when we have a very large series of categories, it becomes a bit complicated to group the means into homogeneous subsets. Would there be a possibility of presenting a table of means with the significant differences grouped into homogeneous subsets to make it easier to identify them?

Thank you very much for your help

[JorgeCamachoS]

Hello

Look at Rafter, Abell & Braselton. 2002. Multiple comparison methods for means. SIAM REVIEW Vol 44. Nº 2, pp. 259-278.

"Fisher’s LSD test is an example of a two-step, step-down procedure. The widely used procedures known as the Newman–Keuls and the Duncan multiple range tests are examples of step-down procedures. These three procedures are not recommended, because they do not protect the FWER."

Happy New Year

[reason180]

(1) I believe that if you want "LSD," that's the same as "No Correction."
(2) jamovi's ANOVA analysis includes lots of options for post hoc tests: "No Correction," "Tukey," Scheffe," "Bonferroni," "Holm." I believe Holm is widely applicable and widely appropriate. In addition, for any set of p values (whether or not they come come from ANOVA), you can use the Rj module to automatically adjust the p values:

# This code computes the Holm-corrected p values for a vector of
# uncorrected p values. (Options besides "holm" are:
# "hochberg", "hommel", "bonferroni", "BH", "BY", and "fdr".)
p.adjust(c(.027, .141, .018, .053), "holm")
# output: [1] 0.081 0.141 0.072 0.106

Hello everyone, I have discovered this software and I am pleasantly surprised. It is very easy to use, with a very simple and beautiful interface.

I have a series of data in different categories that I would like to analyze with ANOVA to see significant differences. However, I am looking for the possibility of carrying out post-hoc tests and I only see the possibility of carrying out the Tukey test.

I would like to know if there is any possibility of carrying out the LSD or DUCAN test, since in some experimental designs they are the ones that best fit.

On the other hand, when presenting the results, when we have a very large series of categories, it becomes a bit complicated to group the means into homogeneous subsets. Would there be a possibility of presenting a table of means with the significant differences grouped into homogeneous subsets to make it easier to identify them?

Thank you very much for your help

[Zeick]

Hello

Look at Rafter, Abell & Braselton. 2002. Multiple comparison methods for means. SIAM REVIEW Vol 44. Nº 2, pp. 259-278.

"Fisher’s LSD test is an example of a two-step, step-down procedure. The widely used procedures known as the Newman–Keuls and the Duncan multiple range tests are examples of step-down procedures. These three procedures are not recommended, because they do not protect the FWER."

Happy New Year

Thank you very much, I will review the paper that you tell me to understand the methods a little better.
On the other hand, I don't understand why these two methods are not recommended because they don't protect "FWER" (I don't understand what this abbreviation is). On the other hand, in agricultural studies a widely used test is SNK (Student Newman Keuls), but I also don't see the way to execute it in jamovi.

(1) I believe that if you want "LSD," that's the same as "No Correction."
(2) jamovi's ANOVA analysis includes lots of options for post hoc tests: "No Correction," "Tukey," Scheffe," "Bonferroni," "Holm." I believe Holm is widely applicable and widely appropriate. In addition, for any set of p values (whether or not they come come from ANOVA), you can use the Rj module to automatically adjust the p values:

# This code computes the Holm-corrected p values for a vector of
# uncorrected p values. (Options besides "holm" are:
# "hochberg", "hommel", "bonferroni", "BH", "BY", and "fdr".)
p.adjust(c(.027, .141, .018, .053), "holm")
# output: [1] 0.081 0.141 0.072 0.106

In principle I do not need correction, simply that the principles of normality and homogeneity of variances are fulfilled.

On the other hand, I'm not very good at R, I don't know how to program in it and that is why discovering this software has been a success.

How could I execute the module in R that you indicate?

Thank you both very much for your help

[reason180]

Hi. What I'm claiming here is that with Fisher's LSD, the only conservative thing you're doing is that you're looking at the ANOVA F results and then only looking at multiple comparisons (with uncorrected p values for the multiple comparisons) if warranted by the F results. Thus, if you want to compute Fisher's LSD, select "uncorrected." "Uncorrected" equals "Fisher's LSD." Also, yes, the code below will in jamovi's Rj module.

# This code computes the Holm-corrected p values for a vector of
# uncorrected p values. (Options besides "holm" are:
# "hochberg", "hommel", "bonferroni", "BH", "BY", and "fdr".)
p.adjust(c(.027, .141, .018, .053), "holm")
# output: [1] 0.081 0.141 0.072 0.106

[JorgeCamachoS]

Dear Zeick,

From Rafter paper "A more pertinent error rate is called the familywise error rate (FWER) or familywise level of signicance. It is the probability of incorrectly rejecting at least one of the null hypotheses that make up the family. The familywise confidence coefficient can be similarly dened. It is the chance that the sampling will result in confidence intervals that simultaneously contain the specied combinations of the means in the family."

Best regards

Jorge Camacho

[Zeick]

Thank you very much to both,

I think I have a little clearer how to run other types of tests in Jamovi.

On the other hand, when I have many hypotheses to compare, it becomes a bit complicated to read the table since it compares the hypotheses one by one. Would there be any possibility of getting the results separated into homogeneous subsets? That is, grouping values that are statistically the same, for example indicating it in the table with letters (a, b, c...) so that the interpretation of the data is more comfortable.

[reason180]

You could "copy" the table and then paste it into a spreadsheet where you could then manually re-arrange the contents however you want.

[Zeick]

I'm asking this in case there is some automated way to organize this data. Imagine we have 10-11 categories and we have to compare one by one. It can be a lot of work.

Copying the data to a spreadsheet is my first option, but the idea is not to spend too much time on this task.

[reason180]

I don't think there's any way to automatically re-arrange the contents using jamovi.