comparing between male and female model

[michelle roth]

Hey everyone

hope you can help me: I`m working on my masterthesis, with latent growth curve modeling.

I have one model for male and one for female. Now I want to compare the Slopes of the two models (if female decrease steeper over time than males). I didn`t manage to write a multilevel model with both gender in it yet, so I have to compare the two models I have. Is there any way to compare if they differ significant in their slopes? Would a simple t-test be appropriate for example? (the structure of the male and female model is the same)

thank you very much

Michelle

[GS Ashiabi]

Hi Michelle, 

There are two options I can think of from the top of my head.

Option 1:

I think you can write a multi-group model with gender in it and the compare the unstandardized coefficients directly

-    1. label the paths differently

o    f1, f2, f3 / m1, m2, m3 (for both females and males respectively

-    2. write code to compare the paths (example: #compare group effects

o    diff_a := f1 – m1

o    diff_b := f2 - m2

o    diff_c := f3 – m3

-    3. include gender in your estimation statement

o    fit <- growth(model, data = growth, se = "bootstrap", bootstrap = 1000,                            group = "Gender")

Option 2:

You can compare the unstandardized coefficients you have. Read the following paper “Raymond Paternoster, Robert Brame, Paul Mazerolle & Alex Piquero (1998) Using the correct statistical test for the equality of regression coefficients. Criminology, 36(4), 859-866.” 

Below is a link to the paper:  http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.458.9930&rep=rep1&type=pdf

I hope this helps.

[michelle roth]

Hi Ashiabi

thank you so much for your answer! Sadly there is not enough time to write the multi-group-model for my thesis (thesis is build up on the two models) but I`ll try it for my own interest!

I`ve just read the paper, thank you for it! The problem is that the sample consists of couples and therefore I should use some test with dependent groups. Do you have any input here to?

thank you so much

Michelle

[Terrence Jorgensen]

Would a simple t-test be appropriate?

If the group were independent, you could calculate a Wald z test by dividing the difference between the parameters by the pooled SE for that difference, which would be the square root of the sum of their squared SEs.

the sample consists of couples and therefore I should use some test with dependent groups. Do you have any input here to? 

In that case, it is absolutely necessary to analyze the data together because the parameters would not be independent (you would need to account for the sampling covariance in the Wald z test, derived from a single model fitted to both dependent groups).  Multilevel is not necessary, just store the data in wide format, where each variable recorded for both members of the couple is stored with a suffix indicating which member it is (e.g., _m and _f for male and female members of heterosexual-only sample of couples).  There is a popular dyadic model called the Actor--Partner Interdependence Model (APIM) easily implemented in SEM.  Here is a recent publication exploring a longitudinal extension, although certainly other extensions are possible:

https://doi.org/10.1080/10705511.2018.1527223

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[michelle roth]

Thank you very much for your answer and inputs!

The thing is, I want to make latent growth curve modeling, not a APIM model. I also checked the APIM, but prefered doing LGM (having 4 timepoints). We`re working on a multilevel model at the moment, but for my thesis I have to write down what I have at the moment and these are only separate models.. therefore I`d need any test to compare these two models (their slopes respectively).

thank you and best regards

Michelle

[Terrence Jorgensen]

The thing is, I want to make latent growth curve modeling, not a APIM model. I also checked the APIM, but prefered doing LGM (having 4 timepoints).

The APIM isn't 1 model, it is a modeling framework.  You can model parallel growth curves for actor and partner, and their growth factors can affect each other.

 

We`re working on a multilevel model at the moment, but for my thesis I have to write down what I have at the moment and these are only separate models.. therefore I`d need any test to compare these two models (their slopes respectively).

Testing that will require modeling both of their slopes at the same time.  You could do this with a multivariate MLM or with parallel growth curves in SEM.