Hello. I am new to R and new to linear mixed effects modeling. I am trying to model some data which has two factors. Each factor has three levels rather than continuous data. Specifically, we measured speech at Test 1, Test 2 and Test 3. We also had three groups of subjects: RepTP, RepNTP and NoRepNTP. I am having a really hard time interpreting this data since all the examples I have seen in the book I am using (Baayen, 2008) either have continuous variables or factors with only two levels. What I find particularly confusing are the interaction terms in the output. The output doesn't present the full interaction (3 X 3) as I would expect with an ANOVA. Instead, it only presents an interaction term for one Test and one Group, presumably comparing it to the reference Test and reference Group. Therefore, it is hard to know what to do with the interactions that aren't significant. In the book, non-significant interactions are dropped from the model. However, in my model, I'm only ever seeing the 2 X 2 interactions, not the full 3 X 3 interaction, so it's not clear what I should do when only two levels of group and two levels of test interact but the third group doesn't. If anyone can assist me in interpreting the output, I would really appreciate it. I may be trying to interpret it too much like an ANOVA where you would be looking for main effects of Test (was there improvement from Test 1 to Test 2), main effects of Group (was one of the Groups better than the other) and the interactions of the two factors (did one Group improve more than another Group from Test 1 to Test 2, for example). I guess another question to pose here is, is it pointless to do an LME analysis with more than two levels of a factor? Is it too much like trying to do an ANOVA? Alternatively, it's possible that what I'm doing is acceptable, I'm just not able to interpret it correctly. I have provided output from my model to hopefully illustrate my question. I'm happy to provide additional information/output if someone is interested in helping me with this problem. Thank you, Laura Linear mixed model fit by REML Formula: PTR ~ Test * Group + (1 | student) Data: ptr AIC BIC logLik deviance REMLdev -625.7 -559.8 323.9 -706.5 -647.7 Random effects: Groups Name Variance Std.Dev. student (Intercept) 0.0010119 0.03181 Residual 0.0457782 0.21396 Number of obs: 2952, groups: studentID, 20 Fixed effects: Estimate Std. Error t value (Intercept) 0.547962 0.016476 33.26 Testtest2 -0.007263 0.015889 -0.46 Testtest1 -0.050653 0.016305 -3.11 GroupNoRepNTP 0.008065 0.022675 0.36 GroupRepNTP -0.018314 0.025483 -0.72 Testtest2:GroupNoRepNTP 0.006073 0.021936 0.28 Testtest1:GroupNoRepNTP 0.013901 0.022613 0.61 Testtest2:GroupRepNTP 0.046684 0.024995 1.87 Testtest1:GroupRepNTP 0.039994 0.025181 1.59 Note: The reference level for Test is Test3. The reference level for Group is RepTP. The interaction p value (after running pvals.fnc with the MCMC) for Testtest2:GroupRepNTP is p = .062 which I'm willing to accept and interpret since speech data with English Language Learners is particularly variable.
Hi Laura, If you want ANOVA output, ask for it! A general strategy that almost always works in R is to fit 2 models, one without the term(s) you want to test, and one with. Then use the anova() function to test them. (models must be nested, and in the lmer() case you need to use REML FALSE). So, try something like this: m1 <- lmer(PTR ~ Test + Group + (1 | student), data=ptr) m2 <- lmer(PTR ~ Test * Group + (1 | student), data=ptr) anova(m1, m2) Best, Ista On Tue, Oct 12, 2010 at 11:59 PM, Laura Halderman <lkh11 at pitt.edu> wrote:> Hello. ?I am new to R and new to linear mixed effects modeling. ?I am trying to model some data which has two factors. ?Each factor has three levels rather than continuous data. ?Specifically, we measured speech at Test 1, Test 2 and Test 3. ?We also had three groups of subjects: RepTP, RepNTP and NoRepNTP. > > I am having a really hard time interpreting this data since all the examples I have seen in the book I am using (Baayen, 2008) either have continuous variables or factors with only two levels. ?What I find particularly confusing are the interaction terms in the output. ?The output doesn't present the full interaction (3 X 3) as I would expect with an ANOVA.Instead, it only presents an interaction term for one Test and one Group, presumably comparing it to the reference Test and reference Group. ?Therefore, it is hard to know what to do with the interactions that aren't significant. ?In the book, non-significant interactions are dropped from the model. ?However, in my model, I'm only ever seeing the 2 X 2 interactions, not the full 3 X 3 interaction, so it's not clear what I should do when only two levels of group and two levels of test interact but the third group doesn't.> > If anyone can assist me in interpreting the output, I would really appreciate it. ?I may be trying to interpret it too much like an ANOVA where you would be looking for main effects of Test (was there improvement from Test 1 to Test 2), main effects of Group (was one of the Groups better than the other) and the interactions of the two factors (did one Group improve more than another Group from Test 1 to Test 2, for example). ?I guess another question to pose here is, is it pointless to do an LME analysis with more than two levels of a factor? ?Is it too much like trying to do an ANOVA? ?Alternatively, it's possible that what I'm doing is acceptable, I'm just not able to interpret it correctly. > > I have provided output from my model to hopefully illustrate my question. ?I'm happy to provide additional information/output if someone is interested in helping me with this problem. > > Thank you, > ?Laura > > Linear mixed model fit by REML > Formula: PTR ~ Test * Group + (1 | student) > ? Data: ptr > AIC ? ? ? ? ? ? BIC ? ? ? ? ? ? logLik ?deviance ? ? ? ?REMLdev > ?-625.7 ? ? ? ? -559.8 ? ? ? ? ?323.9 ? ? ? ? ? -706.5 ? ? ? ? ?-647.7 > Random effects: > ?Groups Name ? ? ? ? ? ?Variance ? ? ? ?Std.Dev. > ?student ? ? ? ?(Intercept) ? ? 0.0010119 ? ? ? 0.03181 > ?Residual ? ? ? ? ? ? ? ? ? ? ? 0.0457782 ? ? ? 0.21396 > Number of obs: 2952, groups: studentID, 20 > > Fixed effects: > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?Estimate ? ? ? ?Std. Error ? ? ?t value > (Intercept) ? ? ? ? ? ? ? ? ? ? 0.547962 ? ? ? ?0.016476 ? ? ? ?33.26 > Testtest2 ? ? ? ? ? ? ? ? ? ? ? -0.007263 ? ? ? 0.015889 ? ? ? ?-0.46 > Testtest1 ? ? ? ? ? ? ? ? ? ? ? -0.050653 ? ? ? 0.016305 ? ? ? ?-3.11 > GroupNoRepNTP ? 0.008065 ? ? ? ?0.022675 ? ? ? ?0.36 > GroupRepNTP ? ? ? ? ? ? -0.018314 ? ? ? 0.025483 ? ? ? ?-0.72 > Testtest2:GroupNoRepNTP ?0.006073 ? 0.021936 ? ?0.28 > Testtest1:GroupNoRepNTP ?0.013901 ? 0.022613 ? ?0.61 > Testtest2:GroupRepNTP ? 0.046684 ? ? ? ?0.024995 ? ? ? ?1.87 > Testtest1:GroupRepNTP ? 0.039994 ? ? ? ?0.025181 ? ? ? ?1.59 > > Note: The reference level for Test is Test3. ?The reference level for Group is RepTP. ?The interaction p value (after running pvals.fnc with the MCMC) for Testtest2:GroupRepNTP is p = .062 which I'm willing to accept and interpret since speech data with English Language Learners is particularly variable. > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >-- Ista Zahn Graduate student University of Rochester Department of Clinical and Social Psychology http://yourpsyche.org
Hi: On Tue, Oct 12, 2010 at 8:59 PM, Laura Halderman <lkh11@pitt.edu> wrote:> Hello. I am new to R and new to linear mixed effects modeling. I am > trying to model some data which has two factors. Each factor has three > levels rather than continuous data. Specifically, we measured speech at > Test 1, Test 2 and Test 3. We also had three groups of subjects: RepTP, > RepNTP and NoRepNTP. >Do you have three groups of subjects, where each subject is tested on three separate occasions? Are the tests meant to be replicates, or is there some other purpose for why they should be represented in the model? Based on this description, it would appear to me that the groups constitute one factor, the students nested within groups another, with three measurements taken on each student. How many students per group?> > I am having a really hard time interpreting this data since all the > examples I have seen in the book I am using (Baayen, 2008) either have > continuous variables or factors with only two levels. What I find > particularly confusing are the interaction terms in the output. The output > doesn't present the full interaction (3 X 3) as I would expect with an > ANOVA. Instead, it only presents an interaction term for one Test and one > Group, presumably comparing it to the reference Test and reference Group. > Therefore, it is hard to know what to do with the interactions that aren't > significant. In the book, non-significant interactions are dropped from the > model. However, in my model, I'm only ever seeing the 2 X 2 interactions, > not the full 3 X 3 interaction, so it's not clear what I should do when only > two levels of group and two levels of test interact but the third group > doesn't. >Let's get the design straight first and the model will work itself out... Dennis> > If anyone can assist me in interpreting the output, I would really > appreciate it. I may be trying to interpret it too much like an ANOVA where > you would be looking for main effects of Test (was there improvement from > Test 1 to Test 2), main effects of Group (was one of the Groups better than > the other) and the interactions of the two factors (did one Group improve > more than another Group from Test 1 to Test 2, for example). I guess > another question to pose here is, is it pointless to do an LME analysis with > more than two levels of a factor? Is it too much like trying to do an > ANOVA? Alternatively, it's possible that what I'm doing is acceptable, I'm > just not able to interpret it correctly. > > I have provided output from my model to hopefully illustrate my question. > I'm happy to provide additional information/output if someone is interested > in helping me with this problem. > > Thank you, > Laura >> > Linear mixed model fit by REML > Formula: PTR ~ Test * Group + (1 | student) > Data: ptr > AIC BIC logLik deviance REMLdev > -625.7 -559.8 323.9 -706.5 -647.7 > Random effects: > Groups Name Variance Std.Dev. > student (Intercept) 0.0010119 0.03181 > Residual 0.0457782 0.21396 > Number of obs: 2952, groups: studentID, 20 > > Fixed effects: > Estimate Std. Error t value > (Intercept) 0.547962 0.016476 33.26 > Testtest2 -0.007263 0.015889 -0.46 > Testtest1 -0.050653 0.016305 -3.11 > GroupNoRepNTP 0.008065 0.022675 0.36 > GroupRepNTP -0.018314 0.025483 -0.72 > Testtest2:GroupNoRepNTP 0.006073 0.021936 0.28 > Testtest1:GroupNoRepNTP 0.013901 0.022613 0.61 > Testtest2:GroupRepNTP 0.046684 0.024995 1.87 > Testtest1:GroupRepNTP 0.039994 0.025181 1.59 > > Note: The reference level for Test is Test3. The reference level for Group > is RepTP. The interaction p value (after running pvals.fnc with the MCMC) > for Testtest2:GroupRepNTP is p = .062 which I'm willing to accept and > interpret since speech data with English Language Learners is particularly > variable. > ______________________________________________ > R-help@r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >[[alternative HTML version deleted]]