I used the mice package in R to perform multiple imputation for my data:
### multiple inputation by chained equations
imp.data <- mice(data, maxit = 5, m = 5, seed = 92385, print = F)
I want to run a logistic regression model after the MI, and predict the outcome based on the model:
model <- with(imp.data, glm(died ~ agecat + female_1 + insurance + mech + transfer +
iss + mxaisbr1 + maxais + cm_chf_1 + cm_mets_1 + cm_liver_1 +
cm_htn_c_1 + cm_bldloss_1 + state, family = binomial))
However, the predict command does not work:
predict(pool(model), type = c('response'))
It would be much simpler if I have the data with imputed values, but the imputation got 5 imputed data sets, making the post estimation complicated.
Any idea ?
Thanks!!
I'm not sure if the imputed models are saved as a data.frame or a matrix..
But, if you convert the model to a data.frame you can plug on the columns of interest to your original frame.
imp.data <- data.frame(imp.data)
original.df$NewImputtedColumn <- imp.data[, 1] # Assuming you want column 1
Now, you can easily keep the copies of the imputted models and still only work with a single data.frame in your predictive models.
This is what I tend to do anyway, might not be the standard way (I'm not sure).
You are only taking the inputted values from 1 of the imputed models, right?
Related
I would like to ask for help with my project. My goal is to get ROC curve from existing logistic regression.
First of all, here is what I'm analyzing.
glm.fit <- glm(Severity_Binary ~ Side + State + Timezone + Temperature.F. + Wind_Chill.F. + Humidity... + Pressure.in. + Visibility.mi. + Wind_Direction + Wind_Speed.mph. + Precipitation.in. + Amenity + Bump + Crossing + Give_Way + Junction + No_Exit + Railway + Station + Stop + Traffic_Calming + Traffic_Signal + Sunrise_Sunset , data = train_data, family = binomial)
glm.probs <- predict(glm.fit,type = "response")
glm.probs = predict(glm.fit, newdata = test_data, type = "response")
glm.pred = ifelse(glm.probs > 0.5, "1", "0")
This part works fine, I am able to show a table of prediction and mean result. But here comes the problem for me, I'm using pROC library, but I am open to use anything else which you can help me with. I'm using test_data with approximately 975 rows, but variable proc has only 3 sensitivities/specificities values.
library(pROC)
proc <- roc(test_data$Severity_Binary,glm.probs)
test_data$sens <- proc$sensitivities[1:975]
test_data$spec <- proc$specificities[1:975]
ggplot(test_data, aes(x=spec, y=sens)) + geom_line()
Here´s what I have as a result:
With Warning message:
Removed 972 row(s) containing missing values (geom_path).
As I found out, proc has only 3 values as I said.
You can't (and shouldn't) assign the sensitivity and specificity to the data. They are summary data and exist in a different dimension than your data.
Specifically, these two lines are wrong and make no sense at all:
test_data$sens <- proc$sensitivities[1:975]
test_data$spec <- proc$specificities[1:975]
Instead you must either save them to a new data.frame, or use some of the existing functions like ggroc:
ggroc(proc)
If you consider what the ROC curve does, there is no reason to expect it to have the same dimensions as your dataframe. It provides summary statistics of your model performance (sensitivity, specificity) evaluated on your dataset for different thresholds in your prediction.
Usually you would expect some more nuance on the curve (more than the 3 datapoints at thresholds -Inf, 0.5, Inf). You can look at the distribution of your glm.probs - this ROC curve indicates that all predictions are either 0 or 1, with very little inbetween (hence only one threshold at 0.5 on your curve). [This could also mean that you unintentially used your binary glm.pred for calculating the ROC curve, and not glm.probs as shown in the question (?)]
This seems to be more an issue with your model than with your code - here an example from a random different dataset, using the same steps you took (glm(..., family = binomial, predict(, type = "response"). This produces a ROC curve with 333 steps for ~1300 datapoints.
PS: (Ingore the fact that this is evaluated on training data, the point is the code looks alright up to the point of generating the ROC curve)
m1 <- glm(survived ~ passengerClass + sex + age, data = dftitanic, family = binomial)
myroc <- roc(dftitanic$survived,predict(m1, dftitanic, type = "response"))
plot(myroc)
I have data from an Experience Sampling Study, which consists of 8140 observations nested in 106 participants. I want to test if there is a mediation, in which I also want to compare the predictors (X1= socialInteraction_tech, X2= socialInteraction_ftf, M = MPEE_int, Y= wellbeing). X1, X2, and M are person-mean centred in order to obtain the within-person effects. To account for the autocorrelation I have fit a model with an ARMA(2,1) structure. We control for time with the variable "obs".
This is the final model including all variables of interest:
fit_mainH1xmy <- lme(fixed = wellbeing ~ 1 + obs # Controls
+ MPEE_int_centred + socialInteraction_tech_centred + socialInteraction_ftf_centred,
random = ~ 1 + obs | ID, correlation = corARMA(form = ~ obs | ID, p = 2, q = 1),
data = file, method = "ML", na.action=na.exclude)
summary(fit_mainH1xmy)
The mediation is partial, as my predictor X still significantly predicts Y after adding M.
However, I can't find a way to calculate c'(cprime), the indirect effect.
I have found the mlma package, but it looks weird and requires me to do transformations to my data.
I have tried melting the data in a long format and using lmer() to fit the model (following https://quantdev.ssri.psu.edu/sites/qdev/files/ILD_Ch07_2017_Within-PersonMedationWithMLM.html), but lmer() does not let me take into account the moving average (MA-part of the ARMA(2,1) structure).
Does anyone know how I could now obtain the indirect effect?
I want to run a stepwise regression in R to choose the best fit model, my code is attached here:
full.modelfixed <- glm(died_ed ~ age_1 + gender + race + insurance + injury + ais + blunt_pen +
comorbid + iss +min_dist + pop_dens_new + age_mdn + male_pct +
pop_wht_pct + pop_blk_pct + unemp_pct + pov_100x_npct +
urban_pct, data = trauma, family = binomial (link = 'logit'), na.action = na.exclude)
reduced.modelfixed <- stepAIC(full.modelfixed, direction = "backward")
There is a error message said
Error in stepAIC(full.modelfixed, direction = "backward") :
number of rows in use has changed: remove missing values?
Almost every variable in the data has some missing values, so I cannot delete all missing values (data = na.omit(data))
Any idea on how to fix this?
Thanks!!
This should probably be in a stats forum (stats.stackexchange) but briefly there are a number of considerations.
The main one is that when comparing two models they need to be fitted on the same dataset (i.e you need to be able to nest the models within each other).
For examples
glm1 <- glm(Dependent~indep1+indep2+indep3, family = binomial, data = data)
glm2 <- glm(Dependent~indep2+indep2, family = binomial, data = data)
Now imagine that we are missing values of indep3 but not indep1 or indep2.
When we run glm1 we are running it on a smaller dataset - the dataset for which we have the dependent variable and all three independent ones (i.e we exclude any rows where indep3 values are missing).
When we run glm2 the rows missing a value for indep3 are included because those rows do contain dependent, indep1 and indep2 which are the models in the variable.
We can no longer directly compare models as they are fitted on different datasets.
I think broadly you can either
1) Limit to data which is complete
2) If appropriate consider multiple imputation
Hope that helps.
You can use the MICE package to do imputation, then working with the dataset will not give you errors
I am using Afrobarometer survey data using 2 rounds of data for 10 countries. My DV is a binary 0-1 variable. I need to use logistic regression, fixed-effects, clustered standard errors (at country), and weighted survey data. A variable for the weights already exists in the dataframe.
I've been looking at help files for the following packages: clogit, glm, pglm, glm2, zelig, bife , etc. Typical errors include: can't add weights, can't do fixed effects, cant do either or etc.
#Glm
t3c1.fixed <- glm(formula = ethnic ~ elec_prox +
elec_comp + round + country, data=afb,
weights = afb$survey_weight,
index c("country", "round"),
family=binomial(link='logit'))
#clogit
t3c1.fixed2 <- clogit(formula = ethnic ~ elec_prox +
elec_comp + round + country, data=afb,
weights = afb$survey_weight,
method=c("within"))
#bife attempt
library(bife)
t3c1.fixed3 <- bife(ethnic ~ elec_prox + elec_comp + round +
country, model = logit,data=afb,
weights = afb$survey_weight,
bias_corr = "ana")
I either get error messages or the code doesn't include one of the conditions I need to include, so I can't use them. In Stata it appears this process is very simple, but in R it seems rather tedious. Any help would be appreciated!
I would check out the survey package which provides everything for which you are asking. The first step is to create the survey object, specify the survey weights and then you are off to the races.
library(survey)
my_survey <- svydesign(ids= ~1, strata = ~country, wts = ~wts, data = your_data)
# Then you can use the survey glm to do what you want via
svy_fit <- svy_glm(ethnic ~ elec_prox +
elec_comp + round + country, data = my_survey, family = binomial())
Or at least I would go down this path given you are using survey data.
I want to fit a linear mixed-effects model using lme4::lmer without discarding observations with missing data. That is, I want lmer to go ahead and maximize the likelihood using all the data.
Am I correct in thinking that using na.pass produces this behavior? This unanswered question is making me wonder if this might be wrong.
lmer(like most model functions) can't deal with missing data. To illustrate that:
data(Orthodont,package="nlme")
Orthodont$nsex <- as.numeric(Orthodont$Sex=="Male")
Orthodont$nsexage <- with(Orthodont, nsex*age)
Orthodont[1, 2] <- NA
lmer(distance ~ age + (age|Subject) + (0+nsex|Subject) +
(0 + nsexage|Subject), data=Orthodont, na.action = na.pass)
#Error in lme4::lFormula(formula = distance ~ age + (age | Subject) + (0 + :
# NA in Z (random-effects model matrix): please use "na.action='na.omit'" or "na.action='na.exclude'"
If you don't want to discard observations with missing data, your only option is imputation. Check out packages like mice or Amelia.