Discover the world's scientific knowledge

Hi,

I have daily level data nested within individuals from a diary study. Let's say I only have hypotheses on between level effects: in this case, is it then always wrong to aggregate the data over the days and use standard OLS type analyses (e.g. correlations)? 

I know there are some concerns with aggregation, e.g. it assumes that within variance is zero, gives people with fewer observations relatively more weight).The latter is less of a problem in my data since people reported about equal numbers of observations and quite a lot of them.

But given that I only have expectations on between person effects, is aggregation such a bad thing?

My research plan is as follows:

5 organisations are taking part in the project. Their employees will get a questionnaire in the beginning, middle and end (t1, t2, t3) of the project.

However, we will not be recording participant data, and so it is not fully longitudinal and more of a cohort study I believe, because we cannot tell whether the same people take part at each time point.

My plan was to do some type of multilevel model with participants nested within organisations, and to measure the effect of time on 3 outcome variables measured using the questionnaire.

Now a reviewer is asking for a sample size calculation to see how many people I would need to recruit for adequate power.

There are so many different programs (free or paid) as well as R packages that can do these types of analyses, and I am not quite sure what to pick. Any advice would be helpful!

In gsem of STATA we can test random-intercept and random-slope models (multilevel) (see example38g in the manual). STATA MULTILEVEL MIXED-EFFECTS "me" deals with multilevel mixed-models, in particular MIXED for continuous outcomes.

I asked myself: Do I get the same results if I use gsem or MIXED? For the moment my answer is yes and no.....

In MIXED we have several options: we can use ML or REML estimation method; we can define different residual variance structures,....

I ran a gsem 2-level random-intercept model (id defines level 2 and session_coded defines level 1 nested within level 2) using own data

gsem (rd <- mpa_level i.session_coded i.order M1[id])

I found out that I get exactly the same results with the following mixed model

mixed rd c.mpa_level i.session_coded i.order ||id:, ml cformat(%9.4f)

However, using reml is prefarable; furthermore, an heterogenous residual variance better fits the specific data rather than the default. So the "best" MIXED model I would use is

mixed rd i.session_coded c.mpa_level i.order ||id:, reml residuals(ind, by(session_coded)) cformat(%9.4f)

With this model, the results are quite different.

Why am I asking this question? Because in a second stage I would like to run multilevel-mediation analyses using gsem but ideally I would like to keep the level of "sophistication" that I have with MIXED (reml, residual variance, etc.).

Best regards,

Hi everyone,

I have a question about how to do a multilevel analysis in SPSS with several mediator variables. I always perform multilevel analyses with the MIXED procedure, but it seems this is a lot more complicated when you want to conduct a mediation analysis. I have two models with one independent variable, 5 mediator variables and one dependent (continuous) variable. Any ideas about how to go about this? Thank you very much.

Dear all,

I am hoping for suggestions on articles/research reports on 3 level multi-level analyses (individuals nested in country-regions (the EU NUTS nomenclature ideally) which are nested in countriees) based on ESS data sets.

Can anyone suggest such type of articles that address the following topics: well-being and life satisfaction, and age groups comparisons?

Thank you all,

Adrian Stanciu

Hi, all.

I am preparing a manuscript and trying to conduct sensitivity test for the multilevel analyses I reported in the manuscript. The software for sensitivity test, Optimal Design, requires specification of effect size Grass' delta. But for the rest of the manuscript, I reported effect size in terms of r. So I am wondering if there is a way to convert Delta into r or some rule of thumb that can help make sense of Delta in r terms? Thank you!

Dear colleagues, I am working with educational data. For this, I am using the classic three-level hierarchical linear model (student, class and school). I'm using Sta version 17 for the analyses. When I perform the residual analysis, the homoscedasticity and normality standards are not met. Here is the adjusted model: xtmixed en_ex_9mat gen_alun rep_alun comp_alun b4.educ_ee gen_prof b1.idad_prof nro_alun_turm b1.sase_esc b1.reg_esc area_esc||id_esc:||id_turm:,mle var Comments: The dependent variable of pt_ex (Exam Score) despite being continuous, has only discrete values ​​(0 to 100). These values ​​are multiples of 4 because the test applied consisted of 25 questions. Regarding the independent variables, with the exception of the variable nro_alun_turm (Number of students in the class), these are nominal/binary categorical. I thought of using a GLMM, ie a Poisson or Negative Binomial multilevel model, but these have infinite support. or could you try a Gamma, since the two assumptions of the Gaussian model are not met? Another question, how do I correct autocorrelated errors at different levels (in this case, at levels 2 and 3, respectively)? Could you please help with this? Thank you very much in advance, Ricardo

Hi all,

We would like to randomize clinics to an intervention arm or usual care arm. We wish to examine difference in outcomes between the two arms as a first step before proceeding to more detailed analyses at the individual level. For the implementation outcome (e.g. Referral to specialty tx), we are interested in differences between the two arms. Can we do a simple difference-in-means analysis with rates summarized at the clinic level, ignoring clustering (patients within providers within clinics)? Thanks.

- Sujaya

Hi all,  

My study examines a main IV-DV relationship and how this is moderated by M1, M2 and M3. DV is measured on a 7-point Likert scale; IV, M1-M3 are dummy variables.

Only the main relationship is significant. For the moderating relationships, the direct relationships are significant – corresponding with earlier studies – but not the interaction terms (IV*M1, IV*M2, and IV*M3). The results are confirmed by supplementary analyses and robustness checks.

One possible explanation for the non-significant interaction terms could be the binary coding of the variables, which in itself restricts the variance in interaction terms. Can you suggest how to get around this limitation for a subsequent paper with the same data? Someone suggested that I use vectors and angular separation. Is this compatible with multilevel data?

Thank you.

Dear all, I am testing a hypothesis: Is there any relationship between independent factors and health behaviour among children’s. My database has two levels: individual (children’s sociodemographic data: age, gender, parents education) and LvL2 variables (schools). I wanted to analyses the differences between the schools also. I guess need to use multilevel regression approach because the children’s individual data are nested into the schools. I have some trouble in the model specification: Outcome: smoking: Yes/No. Independent variables: age, sex, parents’ education level and the schools. In SPSS with the Generalized Linear Mixed Model procedure in the field “Subject” I specified the LvL2 variable which is the school code (schools). In the random effects field I also specified a random intercept (with the variable school code). In the Fixed effects field I specified all independent variables (age, sex, parents’ education level and the also the school code). The analysis results in an error. When I am drop out the school code from the Fixed effects field the program reporting no errors (therefore, I think I misspecified the model). Anny suggestion how can I correctly specify the model? Can I use the same variable in the random effect field, and the fixed effect field at the same time? If not how can I analyses the differences between schools?

Thank you very much for the suggestions.