I have an ordinal dependent variable and trying to use a number of independent variables to predict it. I use R. The function I use is clm in the ordinal package, to perform a cumulative link function with a probit link, to be precise:
I tried the function pR2 in the package pscl to get the pseudo R squared with no success.
How do I get pseudo R squareds with the clm function?
Thanks so much for your help.
There are a variety of pseudo-R^2. I don't like to use any of them because I do not see the results as having a meaning in the real world. They do not estimate effect sizes of any sort and they are not particularly good for statistical inference. Furthermore in situations like this with multiple observations per entity, I think it is debatable which value for "n" (the number of subjects) or the degrees of freedom is appropriate. Some people use McFadden's R^2 which would be relatively easy to calculate, since clm generated a list with one of its values named "logLik". You just need to know that the logLikelihood is only a multiplicative constant (-2) away from the deviance. If one had the model in the first example:
library(ordinal)
data(wine)
fm1 <- clm(rating ~ temp * contact, data = wine)
fm0 <- clm(rating ~ 1, data = wine)
( McF.pR2 <- 1 - fm1$logLik/fm0$logLik )
[1] 0.1668244
I had seen this question on CrossValidated and was hoping to see the more statistically sophisticated participants over there take this one on, but they saw it as a programming question and dumped it over here. Perhaps their opinion of R^2 as a worthwhile measure is as low as mine?
Recommend to use function nagelkerke from rcompanion package to get Pseudo r-squared.
When your predictor or outcome variables are categorical or ordinal, the R-Squared will typically be lower than with truly numeric data. R-squared merely a very weak indicator about model's fit, and you can't choose model based on this.
Related
I am adjusting a mixed effects model which, due to the observed heteroscedasticity, it was necessary to include an effect to accommodate it. Therefore, using the lme function of the nlme package, this was easy to be solved, see the code below:
library(nlme)
library(lme4)
Model1 <- lme(log(Var1)~log(Var2)+log(Var3)+
(Var4)+(Var5),
random = ~1|Var6, Data1, method="REML",
weights = varIdent(form=~1|Var7))
#Var6: It is a factor with several levels.
#Var7: It is a Dummy variable.
However, I need to readjust the model described above using the lme4 package, that is, using the lmer function. It is known and many are the materials that inform some limitations existing in the lme4, such as, for example, modeling heteroscedasticity. What motivated me to readjust this model is the fact that I have an interest in using a specific package that in cases of mixed models it only accepts if they are adjusted through the lmer function. How could I resolve this situation? Below is a good part of the model adjusted using the lmer function, however, this model is not considering the effect to model the observed heteroscedasticity.
Model2 <- lmer(log(Var1)~log(Var2)+log(Var3)+
(Var4)+(Var5)+(1|Var6),
Data1, REML=T)
Regarding the choice of the random effect (Var6) and the inclusion of the effect to consider the heterogeneity by levels of the variable (Var7), these were carefully analyzed, however, I will not put here the whole procedure so as not to be an extensive post and to be more objective .
This is hackable. You need to add an observation-level random effect that is only applied to the group with the larger residual variance (you need to know this in advance!), via (0+dummy(Var7,"1")|obs); this has the effect of multiplying each observation-level random effect value by 1 if the observation is in group "1" of Var7, 0 otherwise. You also need to use lmerControl() to override a few checks that lmer does to try to make sure you are not adding redundant random effects.
Data1$obs <- factor(seq(nrow(Data1)))
Model2 <- lmer(log(Var1)~log(Var2)+log(Var3)+
(Var4)+(Var5) + (1|Var6) +
(0+dummy(Var7,"1")|obs),
Data1, REML=TRUE,
control=lmerControl(check.nobs.vs.nlev="ignore",
check.nobs.vs.nRE="ignore"))
all.equal(REMLcrit(Model2), c(-2*logLik(Model1))) ## TRUE
all.equal(fixef(Model1), fixef(Model2), tolerance=1e-7)
If you want to use this model with hnp you need to work around the fact that hnp doesn't pass the lmerControl option properly.
library(hnp)
d <- function(obj) resid(obj, type="pearson")
s <- function(n, obj) simulate(obj)[[1]]
f <- function(y.) refit(Model2, y.)
hnp(Model2, newclass=TRUE, diagfun=d, simfun=s, fitfun=f)
You might also be interested in the DHARMa package, which does similar simulation-based diagnostics.
I need to perform glm (poisson) estimations with fixed-effects (say merely unit FE) and several regressors (RHS variables). I have an unbalanced panel dataset where most (~90%) observations have missing values (NA) for some but not all regressors.
fixest::feglm() can handle this and returns my fitted model.
However, to do so, it (and fixest::demean too) removes observations that have at least one regressor missing, before constructing the fixed-effect means.
In my case, I am afraid this implies not using a significant share of available information in the data.
Therefore, I would like to demean my variables by hand, to be able to include as much information as possible in each fixed-effect dimension's mean, and then run feglm on the demeaned data. However, this implies getting negative dependent variable values, which is not compatible with Poisson. If I run feglm with "poisson" family and my manually demeaned data, I (coherently) get: "Negative values of the dependent variable are not allowed for the "poisson" family.". The same error is returned with data demeaned with the fixest::demean function.
Question:
How does feglm handle negative values of the demeaned dependent variable? Is there a way (like some data transformation) to reproduce fepois on a fixed-effect in the formula with fepois on demeaned data and a no fixed-effect formula?
To use the example from fixest::demean documentation (with two-way fixed-effects):
data(trade)
base = trade
base$ln_dist = log(base$dist_km)
base$ln_euros = log(base$Euros)
# We center the two variables ln_dist and ln_euros
# on the factors Origin and Destination
X_demean = demean(X = base[, c("ln_dist", "ln_euros")],
fe = base[, c("Origin", "Destination")])
base[, c("ln_dist_dm", "ln_euros_dm")] = X_demean
and I would like to reproduce
est_fe = fepois(ln_euros ~ ln_dist | Origin + Destination, base)
with
est = fepois(ln_euros_dm ~ ln_dist_dm, base)
I think there are two main problems.
Modelling strategy
In general, it is important to be able to formally describe the estimated model.
In this case it wouldn't be possible to write down the model with a single equation, where the fixed-effects are estimated using all the data and other variables only on the non-missing observations. And if the model is not clear, then... maybe it's not a good model.
On the other hand, if your model is well defined, then removing random observations should not change the expectation of the coefficients, only their variance. So again, if your model is well specified, you shouldn't worry too much.
By suggesting that observations with missing values are relevant to estimate the fixed-effects coefficients (or stated differently, that they are used to demean some variables) you are implying that these observations are not randomly distributed. And now you should worry.
Just using these observations to demean the variables wouldn't remove the bias on the estimated coefficients due to the selection to non-missingness. That's a deeper problem that cannot be removed by technical tricks but rather by a profound understanding of the data.
GLM
There is a misunderstanding with GLM. GLM is a super smart trick to estimate maximum likelihood models with OLS (there's a nice description here). It was developed and used at a time when regular optimization techniques were very expensive in terms of computational time, and it was a way to instead employ well developed and fast OLS techniques to perform equivalent estimations.
GLM is an iterative process where typical OLS estimations are performed at each step, the only changes at each iteration concern the weights associated to each observation. Therefore, since it's a regular OLS process, techniques to perform fast OLS estimations with multiple fixed-effects can be leveraged (as is in the fixest package).
So actually, you could do what you want... but only within the OLS step of the GLM algorithm. By no means you should demean the data before running GLM because, well, it makes no sense (the FWL theorem has absolutely no hold here).
I'm relatively new to glm - so please bear with me.
I have created a glm (logistic regression) to predict whether an individual CONTINUES studies ("0") or does NOTCONTINUE ("1"). I am interested in predicting the latter. The glm uses seven factors in the dataset and the confusion matrices are very good for what I need and combining seven years' of data have also been done. Straight-forward.
However, I now need to apply the model to the current years' data, which of course does not have the NOTCONTINUE column in it. Lets say the glm model is "CombinedYears" and the new data is "Data2020"
How can I use the glm model to get predictions of who will ("0") or will NOT ("1") continue their studies? Do I need to insert a NOTCONTINUE column into the latest file ?? I have tried this structure
Predict2020 <- predict(CombinedYears, data.frame(Data2020), type = 'response')
but the output only holds values <0.5.
Any help very gratefully appreciated. Thank you in advance
You mentioned that you already created a prediction model to predict whether a particular student will continue studies or not. You used the glm package and your model name is CombinedYears.
Now, what you have to know is that your problem is a binary classification and you used logistic regression for this. The output of your model when you apply it on new data, or even the same data used to fit the model, is probabilities. These are values between zero and one. In the development phase of your model, you need to determine the cutoff threshold of these probabilities which you can use later on when you predict new data. For example, you may determine 0.5 as a cutoff, and every probability above that is considered NOTCONTINUE and below that is CONTINUE. However, the best threshold can be determined from your data as well by maximizing both specificity and sensitivity. This can be done by calculating the area under the receiver operating characteristic curve (AUC). There are many packages than can do this for you, such as pROC and AUC packages in R. The same packages can determine the best cutoff as well.
What you have to do is the following:
Determine the cutoff threshold after calculating the AUC
library(pROC)
roc_object = roc(your_fit_data$NOTCONTINUE ~ fitted(CombinedYears))
coords(roc.roc_object, "best", ret="threshold", transpose = FALSE)
Use your model to predict on your new data year (as you did)
Predict2020 = predict(CombinedYears, data.frame(Data2020), type = 'response')
Now, the content of Predict2020 is just probabilities for each
student. Use the cutoff you obtained from step (1) to classify your
students accordingly
I am conducting an analysis of where on the landscape a predator encounters potential prey. My response data is binary with an Encounter location = 1 and a Random location = 0 and my independent variables are continuous but have been rescaled.
I originally used a GLM structure
glm_global <- glm(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs,
data=Data_scaled, family=binomial)
but realized that this failed to account for potential spatial-autocorrelation in the data (a spline correlogram showed high residual correlation up to ~1000m).
Correlog_glm_global <- spline.correlog (x = Data_scaled[, "Y"],
y = Data_scaled[, "X"],
z = residuals(glm_global,
type = "pearson"), xmax = 1000)
I attempted to account for this by implementing a GLMM (in lme4) with the predator group as the random effect.
glmm_global <- glmer(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs+(1|Group),
data=Data_scaled, family=binomial)
When comparing AIC of the global GLMM (1144.7) to the global GLM (1149.2) I get a Delta AIC value >2 which suggests that the GLMM fits the data better. However I am still getting essentially the same correlation in the residuals, as shown on the spline correlogram for the GLMM model).
Correlog_glmm_global <- spline.correlog (x = Data_scaled[, "Y"],
y = Data_scaled[, "X"],
z = residuals(glmm_global,
type = "pearson"), xmax = 10000)
I also tried explicitly including the Lat*Long of all the locations as an independent variable but results are the same.
After reading up on options, I tried running Generalized Estimating Equations (GEEs) in “geepack” thinking this would allow me more flexibility with regards to explicitly defining the correlation structure (as in GLS models for normally distributed response data) instead of being limited to compound symmetry (which is what we get with GLMM). However I realized that my data still demanded the use of compound symmetry (or “exchangeable” in geepack) since I didn’t have temporal sequence in the data. When I ran the global model
gee_global <- geeglm(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs,
id=Pride, corstr="exchangeable", data=Data_scaled, family=binomial)
(using scaled or unscaled data made no difference so this is with scaled data for consistency)
suddenly none of my covariates were significant. However, being a novice with GEE modelling I don’t know a) if this is a valid approach for this data or b) whether this has even accounted for the residual autocorrelation that has been evident throughout.
I would be most appreciative for some constructive feedback as to 1) which direction to go once I realized that the GLMM model (with predator group as a random effect) still showed spatially autocorrelated Pearson residuals (up to ~1000m), 2) if indeed GEE models make sense at this point and 3) if I have missed something in my GEE modelling. Many thanks.
Taking the spatial autocorrelation into account in your model can be done is many ways. I will restrain my response to R main packages that deal with random effects.
First, you could go with the package nlme, and specify a correlation structure in your residuals (many are available : corGaus, corLin, CorSpher ...). You should try many of them and keep the best model. In this case the spatial autocorrelation in considered as continous and could be approximated by a global function.
Second, you could go with the package mgcv, and add a bivariate spline (spatial coordinates) to your model. This way, you could capture a spatial pattern and even map it. In a strict sens, this method doesn't take into account the spatial autocorrelation, but it may solve the problem. If the space is discret in your case, you could go with a random markov field smooth. This website is very helpfull to find some examples : https://www.fromthebottomoftheheap.net
Third, you could go with the package brms. This allows you to specify very complex models with other correlation structure in your residuals (CAR and SAR). The package use a bayesian approach.
I hope this help. Good luck
I have been doing variable selection for a modeling problem.
I have used trial and error for the selection (adding / removing a variable) with a decrease in error. However, I have the challenge as the number of variables grows into the hundreds that manual variable selection can not be performed as the model takes 1/2 hour to compute, rendering the task impossible.
Would you happen to know of any other packages than the regsubsets from the leaps package (which when tested with the same trial and error variables produced a higher error, it did not include some variables which were lineraly dependant - excluding some valuable variables).
You need a better (i.e. not flawed) approach to model selection. There are plenty of options, but one that should be easy to adapt to your situation would be using some form of regularization, such as the Lasso or the elastic net. These apply shrinkage to the sizes of the coefficients; if a coefficient is shrunk from its least squares solution to zero, that variable is removed from the model. The resulting model coefficients are slightly biased but they have lower variance than the selected OLS terms.
Take a look at the lars, glmnet, and penalized packages
Try using the stepAIC function of the MASS package.
Here is a really minimal example:
library(MASS)
data(swiss)
str(swiss)
lm <- lm(Fertility ~ ., data = swiss)
lm$coefficients
## (Intercept) Agriculture Examination Education Catholic
## 66.9151817 -0.1721140 -0.2580082 -0.8709401 0.1041153
## Infant.Mortality
## 1.0770481
st1 <- stepAIC(lm, direction = "both")
st2 <- stepAIC(lm, direction = "forward")
st3 <- stepAIC(lm, direction = "backward")
summary(st1)
summary(st2)
summary(st3)
You should try the 3 directions and ckeck which model works better with your test data.
Read ?stepAIC and take a look at the examples.
EDIT
It's true stepwise regression isn't the greatest method. As it's mentioned in GavinSimpson answer, lasso regression is a better/much more efficient method. It's much faster than stepwise regression and will work with large datasets.
Check out the glmnet package vignette:
http://www.stanford.edu/~hastie/glmnet/glmnet_alpha.html