Using the 1974 Motor Trend mtcars data (32
cars), we asked whether manual or automatic transmission is better for
fuel economy (MPG) and how large the difference is. A naive comparison
shows manual-transmission cars average 7.24 MPG more
than automatic cars (24.4 vs. 17.1 MPG, Welch t = -3.77, p = 0.0014).
However, manual cars in this sample are also systematically lighter and
quicker, and both traits are strongly associated with MPG. After
adjusting for vehicle weight and quarter-mile
time, transmission type still matters, but the effect shrinks
to +2.94 MPG for manual over automatic (95% CI: 0.05 to
5.83 MPG, p = 0.047) — an estimate of the effect holding weight and
acceleration fixed, not the raw group difference. Weight is the dominant
predictor: each additional 1,000 lb costs about 3.92 MPG. We conclude
manual transmission is associated with better mileage, but most of the
raw 7.24 MPG gap is explained by manual cars being lighter, not by the
transmission itself.
The boxplot (Figure 1) shows manual cars have both a higher median
and higher spread of MPG. Pairwise correlations (Appendix, Fig. A1) show
wt (weight, r = -0.87) and qsec (1/4-mile
time, r = 0.42) are the variables most associated with mpg
besides am, and wt and am are
themselves correlated (manual cars tend to be lighter) — a classic
confounding pattern that motivates a multivariable
model rather than the simple two-group comparison.
We began with the simple regression mpg ~ am, which
simply reproduces the raw group means and confirms the unadjusted 7.24
MPG gap (R² = 0.36 — transmission alone explains only 36% of MPG
variation). We then fit the full model with all 10
other variables, which raised R² to 0.88 but left am
statistically insignificant (p = 0.21) with badly inflated standard
errors — a sign of multicollinearity among the 10 correlated predictors
(e.g., cyl, disp, hp,
wt are all measuring “bigger/heavier car”).
To find a parsimonious model we used stepwise selection (both
directions, AIC) starting from the full model, which converged on: \[\text{mpg} = \beta_0 + \beta_1\,\text{am} +
\beta_2\,\text{wt} + \beta_3\,\text{qsec}\] An ANOVA comparing
this model against the am-only model confirms
wt and qsec jointly add significant
explanatory power (F = 45.6, p < 0.001), while adding the remaining 8
variables back does not significantly improve fit over
this 3-variable model (F = 0.88, p = 0.55). Variance inflation factors
for wt and qsec are low (2.48 and 1.36),
indicating no problematic collinearity remains. We select this as our
final model.
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 9.618 | 6.960 | 1.382 | 0.178 |
| am_fManual | 2.936 | 1.411 | 2.081 | 0.047 |
| wt | -3.917 | 0.711 | -5.507 | 0.000 |
| qsec | 1.226 | 0.289 | 4.247 | 0.000 |
Model fit: R² = 0.85 (adjusted 0.83); the model explains 85% of the variability in MPG using only transmission type, weight, and quarter-mile time.
Is manual or automatic better for MPG? Manual transmission is associated with higher MPG, both unadjusted and after controlling for weight and acceleration. The adjusted effect remains statistically significant at the 5% level (p = 0.047).
Quantifying the difference: Holding weight and quarter-mile time constant, switching from automatic to manual transmission is associated with an increase of 2.94 MPG (95% CI: 0.05 to 5.83 MPG). By contrast, the unadjusted difference was 7.24 MPG — most of that gap is attributable to manual cars being lighter (each 1,000 lb of weight costs 3.92 MPG, 95% CI: -5.37 to -2.46, p < 0.001) and having faster quarter-mile times, not to the transmission mechanism itself. As a sensitivity check we also fit a model allowing transmission and weight to interact (Appendix C); it shows the manual-transmission advantage is largest for light cars and essentially disappears for the heaviest cars, reinforcing that weight — not transmission per se — drives most of the fuel-economy difference.
Residuals vs. fitted values show no strong pattern, supporting linearity and roughly constant variance. The Normal Q-Q plot is close to the line (a Shapiro-Wilk test on the residuals gives p = 0.08, failing to reject normality). No point has Cook’s distance near 1 (max = 0.35, for the Chrysler Imperial), so no single car unduly drives the fit, though the Imperial and Merc 230 have moderately high leverage and are worth flagging as outliers of interest in future data collection.
After adjusting for confounding by vehicle weight and acceleration, manual transmission cars in this sample get significantly better mileage than automatics — about 2.94 MPG more, holding weight and quarter-mile time fixed (95% CI 0.05–5.83 MPG) — but the bulk of the commonly-cited automatic-vs-manual MPG gap is really a weight effect: manual cars in this data happen to be substantially lighter.
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 17.820 | 16.306 | 1.093 | 0.287 |
| am_fManual | 2.617 | 2.005 | 1.306 | 0.207 |
| wt | -3.806 | 1.847 | -2.061 | 0.053 |
| hp | -0.046 | 0.027 | -1.701 | 0.104 |
| cyl_f6 | -1.660 | 2.262 | -0.734 | 0.472 |
| cyl_f8 | 1.637 | 4.316 | 0.379 | 0.708 |
| disp | 0.014 | 0.017 | 0.799 | 0.433 |
| drat | 0.026 | 1.676 | 0.016 | 0.988 |
| qsec | 0.647 | 0.722 | 0.896 | 0.381 |
| vs | 1.747 | 2.273 | 0.769 | 0.451 |
| gear | 0.764 | 1.457 | 0.525 | 0.606 |
| carb | 0.509 | 0.942 | 0.540 | 0.595 |
| Res.Df | RSS | Df | Sum of Sq | F | Pr(>F) |
|---|---|---|---|---|---|
| 30 | 720.8966 | NA | NA | NA | NA |
| 28 | 169.2859 | 2 | 551.61067 | 41.3738666 | 0.0000001 |
| 20 | 133.3235 | 8 | 35.96248 | 0.6743464 | 0.7083175 |
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 9.723 | 5.899 | 1.648 | 0.111 |
| am_fManual | 14.079 | 3.435 | 4.099 | 0.000 |
| wt | -2.937 | 0.666 | -4.409 | 0.000 |
| qsec | 1.017 | 0.252 | 4.035 | 0.000 |
| am_fManual:wt | -4.141 | 1.197 | -3.460 | 0.002 |
The significant negative interaction term (am:wt, p = 0.002) means the manual-transmission MPG advantage declines as weight increases: at low weights manual cars have a large advantage, but for very heavy cars the predicted advantage shrinks toward zero. This is consistent with the main model’s finding that weight is the primary driver of MPG differences.
| Group | Mean_MPG | SD_MPG | N.Var1 | N.Freq | |
|---|---|---|---|---|---|
| Automatic | Automatic | 17.15 | 3.83 | Automatic | 19 |
| Manual | Manual | 24.39 | 6.17 | Manual | 13 |