0

I have a data frame called mxDF looks like this:

> head(mxDF)
     y x.Age x.Weight x.Height x.Medicine x.genotype1
1 3.00    36       72      171                  0                    0
2 1.50    63       49      162                  0                    0
3 2.25    40       44      154                  0                    0
4 2.25    57       41      159                  0                    0
5 2.25    59       55      157                  0                    0
6 1.50    45       51      160                  0                    0
  x.genotype2
1                      0
2                      0

Then I did the following:

mxMod     <- glm(mxDF)
res.mxMod <- resid(mxMod);
plot(res.mxMod ~ mxDF[,1]);
abline(glm(res.mxMod ~ mxDF[,1]))

And the plot looks like below:

enter image description here

The pattern on the residue plot is so obvious, and then I did a glm on the residue ~ outcome:

summary(glm(res.mxMod ~ mxDF[,1]))
Call:
glm(formula = res.mxMod ~ mxDF[, 1])

Deviance Residuals: 
     Min        1Q    Median        3Q       Max  
-3.04479  -0.28613   0.03926   0.36334   2.34211  

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept) -1.54024    0.04782  -32.21   <2e-16 ***
mxDF[, 1]    0.57049    0.01630   34.99   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for gaussian family taken to be 0.3229295)

    Null deviance: 693.21  on 923  degrees of freedom
Residual deviance: 297.74  on 922  degrees of freedom
AIC: 1581.8

Number of Fisher Scoring iterations: 2

Isn't this weird? What did I do wrong?

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Mousheng Xu
  • 71
  • 8
  • What is `glm(mxDF)`? I have no idea what that is supposed to be. What model were you hoping to fit? Note that this looks more like R help than a statistical question; this may be off topic here. – gung - Reinstate Monica Oct 17 '16 at 16:11
  • 1
    Perhaps plot the residuals against the predicted rather than the observed and then have a re-think? – mdewey Oct 17 '16 at 16:13
  • Sorry, I am using R, glm is generalized linear model, and mxDF is a data frame with y & predictors as "columns". – Mousheng Xu Oct 17 '16 at 17:21
  • 2
    Useful question about correlation between residual and dependent variable: http://stats.stackexchange.com/questions/5235/what-is-the-expected-correlation-between-residual-and-the-dependent-variable – Pieter Oct 17 '16 at 18:11
  • @mdewey OK, you are right. I should plot residual ~ predicted instead of observed. Thanks. – Mousheng Xu Oct 17 '16 at 18:26
  • Can you please add your coments as an answer so the Q do not stand as unanswered? – kjetil b halvorsen Sep 16 '17 at 16:14

0 Answers0