First of all I recommend using the newest version of lsmeans which is called emmeans.

Bonferroni correction can be aplied by using 'adjust = "bonferroni"' but I do not think that Bonferroni correction is reasonable to apply as it is to conservative. i always found the default adjustement options most suitable for my cases, maybe you should check whether the requirements for theses adjustements are fulfilled by your data.

Concerning the comparison of some factor levels, instead of using pairwise comparison you could use the 'trt.vs.ctrl' comparison:

contrast(YOUREMMEANSOBJECT, "trt.vs.ctrl", ref = "STANDARD")

Sven Greving thank you very much for your answer!

It looks beautiful with with function contrust(). Thank you :-)

How can I check whether the default adjustment is suitable for my data?

Best regards

Anna

Glad it helps, check out the vignette for emmeans adjustment methods

Dear Sven Greving

Can you help me with a little thing regarding post hoc using emmeans? I am sorry for so long post, but it is really a little thing, I just want to explain the problem in details.

I used the mixed effect model to analyze my data. Here is the model:

Model

The only thing I can think of, that might cause this problem is the coding of the MeasureSide variable. How is it coded (dummy, contrast, effect...)and is the coding correct in your dataframe? Could you please provide the summary for 'Model'?

Sven Greving, thank you for your fast response. Here is the summary:

Linear mixed model fit by REML ['lmerMod']

Formula: Amp ~ Level + Stim + Level:Stim + Threshold + MeasureSide + 0 + (1 | Subject) + (1 | Subject:Ear)

Data: D

REML criterion at convergence: 1890.3

Scaled residuals:

Min 1Q Median 3Q Max

-4.0790 -0.6254 0.0837 0.6403 2.8029

Random effects:

Groups Name Variance Std.Dev.

Subject:Ear (Intercept) 0.635 0.7969

Subject (Intercept) 7.329 2.7073

Residual 2.675 1.6355

Number of obs: 480, groups: Subject:Ear, 16; Subject, 10

Fixed effects:

Estimate Std. Error t value

Level30dBSL 40.70040 1.06312 38.284

Level40dBSL 43.03809 1.06312 40.483

Level50dBSL 45.29444 1.06312 42.605

StimType2_1/3oct 1.04401 0.41668 2.506

StimType3_1/2oct 1.59529 0.42429 3.760

StimType4_1oct 2.95862 0.41668 7.100

StimType5_2oct 4.85626 0.43195 11.243

Threshold 0.09968 0.03777 2.639

MeasureSideIL 0.30193 0.14930 2.022

Level40dBSL:StimType2_1/3oct -0.69254 0.57825 -1.198

Level50dBSL:StimType2_1/3oct -1.76756 0.57825 -3.057

Level40dBSL:StimType3_1/2oct -0.53084 0.57825 -0.918

Level50dBSL:StimType3_1/2oct -1.98681 0.57825 -3.436

Level40dBSL:StimType4_1oct -0.53900 0.57825 -0.932

Level50dBSL:StimType4_1oct -1.70241 0.57825 -2.944

Level40dBSL:StimType5_2oct -1.64722 0.57825 -2.849

Level50dBSL:StimType5_2oct -2.57799 0.57825 -4.458

Best

Anna

Sven Greving I wonder actually why this fixed effect became significant, as I have never experienced that in my data and there is also a lot of literature that reported no difference between IL and CL. I also compared difference between IL and CL with two-sided paired t-test for different stimuli types and intensities and tested showed no difference for any of the stimuli types and intensities (p-values were larger than 0.15). When I test the reduction of the model I do the following: I fit the full model:

lme

Hve you tried direct comparison of MeasureSide in your emmeans? Maybe it is something in the emmeans code

Sven Greving

To be honest I do not know how to do that. Can you help and explain how to investigate that?

Best regards

Anna

Instead of

lme.emm

Sven Greving

I tried this. It gave absolutely the same results for all types of Level:Stim combination. Example is showed below

Level = 30dBSL, Stim = Type1_PT:

contrast estimate SE df t.ratio p.value

CL - IL -0.302 0.149 448 -2.022 0.0437

And this was for all results :-(

If you have any other suggestions, I will appreciate that.

Best regards

Anna