Running speedlm on weighted data with missing values

Question

I am trying to run a linear regression on weighted data.
When using speedlm i get an error msg when there are missing values in the data.

 library(speedglm)
 sampleData <- data.frame(w = round(runif(12,0,1)),
                          target = rnorm(12,100,50),
                          predictor = c(NA, rnorm(10, 40, 10),NA))

 summary(sampleData)

       w              target          predictor    
 Min.   :0.0000   Min.   : -3.381   Min.   :22.58  
 1st Qu.:0.0000   1st Qu.: 48.321   1st Qu.:30.45  
 Median :1.0000   Median : 84.156   Median :37.09  
 Mean   :0.5833   Mean   : 92.306   Mean   :35.03  
 3rd Qu.:1.0000   3rd Qu.:119.891   3rd Qu.:41.96  
 Max.   :1.0000   Max.   :223.896   Max.   :43.48  
                                    NA's   :2

 #run linear regression without weights
 linearNoWeights <- lm(formula("target~predictor"), data = sampleData)
 speedLinearNoWeights <- speedlm(formula("target~predictor"), data = sampleData)

 #run linear regression with weights
 linearWithWeights <- lm(formula("target~predictor"), data = sampleData, weights =sampleData[,"w"] )
 speedLinearWithWheights <- speedlm(formula("target~predictor"), data = sampleData, weights =sampleData[,"w"] )

Error in base::crossprod(x, y) : non-conformable arguments
In addition: Warning messages:
1: In sqw * X :
  longer object length is not a multiple of shorter object length
2: In sqw * y :
  longer object length is not a multiple of shorter object length
Called from: base::crossprod(x, y)

Is there any way around this that does not force me to fix the data before running the regression?

Why are you opposed to removing these two observations from the dataset prior to fitting the model? — Roland, Nov 22 '16 at 08:19
@Roland what I showed here is an example I have actually many data frames and the NA's are important for the rest of the calculations — eliavs, Nov 22 '16 at 08:21

score 1 · Answer 1 · answered Nov 22 '16 at 08:47

You should try to change the na.action option. Below is your code, which I am able to run, when I change na.action to na.exclude/na.omit.

library(speedglm)
sampleData <- data.frame(w = round(runif(12,0,1)),
                         target = rnorm(12,100,50),
                         predictor = c(NA, rnorm(10, 40, 10),NA))
summary(sampleData)

linearNoWeights <- lm(formula("target~predictor"), data = sampleData)
speedLinearNoWeights <- speedlm(formula("target~predictor"), data = sampleData)

options(na.action="na.exclude") # or "na.omit"

linearNoWeights <- lm(formula("target~predictor"), data = sampleData)
    speedLinearNoWeights <- speedlm(formula("target~predictor"), data = sampleData)

You can go through the documentation for na.omit or na.exclude to understand when to use what. Hope this helps.

Running speedlm on weighted data with missing values

1 Answers1