Home / Questions / 349053
Accepted
Juan Luis
Asked: 2020-04-25 04:24:23 +0800 CST 2020-04-25 04:24:23 +0800 CST

How can I get the dataframe resulting from estimating a binomial logit model with glm?

  • 772

I have a dataframewith several missing values ​​and I want to perform an analysis of variance (ANOVA) to compare two binomial logit models :

  • Model A : contains a set of variables.
  • Model B : Contains the same variables as Model A plus 3 study variables .

We import the data:

Model A:

modelo_logit <- glm(SAP ~ sexo + edad + peso + niv_est + enf_cron + sit_lab + 
                       frec_act_fis + ingreso_eq + GHQ_12, 
                       data = datos_modelo, family = binomial(link = "logit"),na.action = "na.omit")

Model B:

modelo_logit_viv <- glm( SAP ~ sexo + edad + niv_est + enf_cron + sit_lab +
                                  frec_act_fis + ingreso_eq + GHQ_12 + 
                                  n_dormitorios + cont_indus + delincuencia, # variables de estudio 
                        data = datos_modelo, family = binomial(link = "logit"),na.action = "na.omit")

In order to perform an ANOVA , I execute:anova(modelo_logit,modelo_logit_viv)

And I get the following error:

Error in anova.glmlist(c(list(object), dotargs), dispersion = dispersion, : models were not all fitted to the same size of dataset

Both models should be fitted by the same for the same data set , but since there are several missing values, model B has more NA'sthan model A (since model B contains more variables which increases the number of observations to be removed). compared to model A).

My question then is: How can I obtain the dataframe used in model B ( modelo_logit_viv) to be able to estimate model A ( modelo_logit) and thus have the same dataframe to estimate both models and then carry out the ANOVA? There must be some element inside the object generated by glm(*)that contains the dataframeone that has been used to estimate once the NA's, but I can't find it.

r

1 Answers

  1. Best Answer

    The element you are looking for is called model, here is an example copied from the help of glm:

    set.seed(2020)
counts <- c(18,17,15,20,10,20,25,13,12)
outcome <- gl(3,1,9)
treatment <- gl(3,3)
random_with_na <- sample(c(1, 2, NA), 9, replace = TRUE)
d.AD <- data.frame(treatment, outcome, counts, random_with_na)

m1 <- glm(counts ~ outcome + treatment, data = d.AD, family = poisson(), na.action = "na.omit")
m2 <- glm(counts ~ outcome + treatment + random_with_na,  data = d.AD, family = poisson(), na.action = "na.omit")

m1$model
m2$model

> m1$model
  counts outcome treatment
1     18       1         1
2     17       2         1
3     15       3         1
4     20       1         2
5     10       2         2
6     20       3         2
7     25       1         3
8     13       2         3
9     12       3         3
> m2$model
  counts outcome treatment random_with_na
1     18       1         1              2
2     17       2         1              1
3     15       3         1              2
4     20       1         2              2
5     10       2         2              2

    As can be seen, the data of model 2 (which adds a new variable with NA's) has fewer rows than the original data, because precisely in this case the complete rows are omitted.

    Now, regarding what you are trying to do, I understand that you have only two possible paths to follow:

    1. Impute the valuesNA
    2. Generate data.frameonly the rows that do not have NAany of the variables of interest, for example, na.ommit(tu_dataframe[, columnas_para_los_modelos])and the data.frameresult would be the one that would feed the two models.
    • 1