Final models evaluation

R For SAS Users

Melinda Higgins, PhD

Research Professor/Senior Biostatistician Emory University

Final exercises

Course wrap-up

Run regression models
- for different predictors
- for different groups
Choose best models
- save model results
- extract and display fit statistics

Showcase your skills

Evaluate and compare models
- using graphical visualizations
Report best associations
- between variables
- overall and by group

Comparing models

# Run lm() for diffht by bmi, save model
lmdiffhtbmi <- lm(diffht ~ bmi,
                  data = daviskeep)

# Run lm() for diffht by weight, save model
lmdiffhtwt <- lm(diffht ~ weight,
                 data = daviskeep)

# Run summary() for each model, save results
smrylmdiffhtbmi <- summary(lmdiffhtbmi)
smrylmdiffhtwt <- summary(lmdiffhtwt)

Comparing models

# Display r.squared for weight model
smrylmdiffhtwt$r.squared

# Display r.squared for bmi model
smrylmdiffhtbmi$r.squared

# Compare AICs for both models
AIC(lmdiffhtbmi, lmdiffhtwt)


[1] 0.003281645


[1] 0.00121824

            df      AIC
lmdiffhtbmi  3 788.0816
lmdiffhtwt   3 787.7052

Models by group - men vs women

# Plot diffht by weight by sex
ggplot(daviskeep,
       aes(diffht, weight)) +
  geom_point() +
  geom_smooth(method = "lm") +
  facet_wrap(vars(sex)) +
  ggtitle("Height differences
          predicted by weight,
          model fit by sex")

plot of diffht by weight by sex

Regression on subset

sas proc reg where option like subset option for R lm function

Fit models for subsets

# lm() of diffht by weight for females
lmdiffhtwtF <- lm(diffht ~ weight,
                  subset = (sex == "F"),
                  data = daviskeep)

# lm() of diffht by weight for males
lmdiffhtwtM <- lm(diffht ~ weight,
                  subset = (sex == "M"),
                  data = daviskeep)

# Run summary() for each model save results
smrylmdiffhtwtF <- summary(lmdiffhtwtF)
smrylmdiffhtwtM <- summary(lmdiffhtwtM)

Fit models for subsets

# r.squared for females only model
smrylmdiffhtwtF$r.squared

# r.squared for males only model
smrylmdiffhtwtM$r.squared


[1] 4.00807e-05


[1] 0.00804139

Let's wrap up by developing a few models for predicting abalone ages!

R For SAS Users