Can MplusAutomation extract multinomial logistic regression coefficients following LCA (R3STEP)?

[Ewan]

Does anyone know if MplusAutomation can extract coefficients from the multinomial logistic regressions carried out as part of a LCA model? (using the R3STEP options)

E.g. the output looks like this:

TESTS OF CATEGORICAL LATENT VARIABLE MULTINOMIAL LOGISTIC REGRESSIONS USING
THE 3-STEP PROCEDURE

                                                    Two-Tailed
                    Estimate       S.E.  Est./S.E.    P-Value

 #1       ON
    AGE               -0.099      0.039     -2.551      0.011
    FEMALE            -0.364      0.178     -2.048      0.041

 #2       ON
    AGE               -0.135      0.047     -2.867      0.004
    FEMALE            -0.346      0.198     -1.751      0.080

 Intercepts
    #1                 1.259      0.419      3.008      0.003
    #2                 2.074      0.502      4.128      0.000

Parameterization using Reference Class 1

[...]

I couldn't find these numbers anywhere the output of readModels. If not, I can go ahead and extract them manually, but thought I'd check first.

Thanks as always, for creating such a useful package!

Ewan

[Ray Sin]

bump... anyone?

[David Allsop]

I would love if this was a feature!

[Simon Brauer]

I was just working on this issue last week. I couldn't find a way to do this within the package, so I wrote my own code to scrape it from the `.out` file. I've included my code below. It uses tidyverse (stringr, dplyr, readr, tidyr, purrr). The main function is `extract_r3step()` which takes the path to your output file. e.g.

    `extract_r3step("output/model1.out")`

I can't claim how generalizable my code is. It certainly isn't pretty. But hope it helps. You may have to edit the line that says "row2 = (1:n())[str_detect(x, 'CONFIDENCE INTERVALS OF MODEL RESULTS')" to whatever text follows the R3STEP section in your output.

```

parse_predictors <- function(x){

  x <- x %>%

    str_replace('^[\\s]+', '') %>%

    str_replace('[\\s]+$', '') %>%

    str_split('[\\s]+') %>%

    unlist()

  

  if(length(x) < 5){

    x <- c(x, rep('', times = 5 - length(x)))

  }

  

  tibble(colnames = c('predictor', 'estimate', 'se', 'tval', 'pval'),

         value = x) %>%

    spread(colnames, value) %>%

    select(predictor, estimate, se, tval, pval)

}

extract_r3step <- function(model_path){

  tibble(x = read_lines(model_path)) %>%

    mutate(row1 = (1:n())[str_detect(x, 'TESTS OF CATEGORICAL LATENT VARIABLE')],

           row2 = (1:n())[str_detect(x, 'CONFIDENCE INTERVALS OF MODEL RESULTS')]) %>%

    slice((.$row1[1] + 2):(.$row2[1] - 1)) %>%

    mutate(type_comparison = str_detect(x, 'Parameterization using Reference'),

           type_colname = str_detect(x, 'Estimate'),

           type_classnum = str_detect(x, 'C#[0-9]+'),

           type_predictor = str_detect(x, '[\\.\\-0-9]+[\\s]+[\\.\\-0-9]+')) %>%

    mutate(class = ifelse(type_classnum, parse_number(x), NA),

           predictors = ifelse(type_predictor, map(x, parse_predictors), list()),

           comparison_group = cumsum(type_comparison)) %>%

    group_by(comparison_group) %>%

    mutate(class_group = cumsum(type_classnum)) %>%

    ungroup() %>%

    mutate(comparison_group = ifelse(comparison_group == 0, max(class, na.rm = TRUE), comparison_group)) %>%

    group_by(comparison_group, class_group) %>%

    filter(class_group != 0) %>%

    mutate(class_group2 = class[!is.na(class)]) %>%

    ungroup() %>%

    filter(type_predictor) %>%

    select(comparison_group, class_group2, predictors) %>%

    unnest() %>%

    rename(comparison_class = comparison_group,

           class = class_group2) %>%

    mutate(estimate = as.numeric(estimate),

           se = as.numeric(se),

           tval = as.numeric(tval),

           pval = as.numeric(pval)) 

}
```

[David Allsop]

Awesome - thanks for sharing!