12
a)
b)
c)
d)
e,f)
S is the standard error of estimate
R² is the coefficient of determination
R² indicates the percentage of the variability in direct labor cost
explained by the number of copies produced.
g) We have H0: B1=0
P value < 0.05 => reject h0, slope is significant
h) Point Forecast Y(250)= 0.9426 + 0.004715x250= 2.1214
90% Prediction Interval
Lower bound = 1.424
Upper bound = 2.818
The point forecast of the direct labor cost for 250 copies is 2.121.
The 90% prediction interval is [1.424, 2.818]. This means that with
90% confidence, the labor cost for a project involving 250 copies is
expected to fall within this range.
i)
The residual plots show that most residuals are centered around
zero, indicating that the linear structure is appropriate.
However, the residuals exhibit mild heteroscedasticity, slight right-
skewness, and several outliers, as seen in the histogram and normal
probability plot. Overall, a simple linear regression model is
reasonably appropriate for these data and provides meaningful
predictions, although the model is not perfect due to non-normality
and unequal variance in the residuals.
13)
a)
b,c)
Nếu p < 0.05 → slope significant → batch size ảnh hưởng số
defecives.
d)
e)
The model explains a large portion of the variability in the number of
defectives (R² = 0.921), and the fitted curve captures the
decreasing marginal increase in defectives as batch size grows.
f)
Null hypothesis: H0:β1=0
Alternative hypothesis: H1:β1≠0
From the regression output:
t-value = 5.559
p-value = 0.000
Because the p-value is far below 0.05, we reject the null
hypothesis.
Conclusion: The slope is statistically significant. The log-
transformed batch size has a meaningful effect on the number of
defectives.
g)
When viewing residual plots (Residuals vs Fits), the residuals do not
appear randomly scattered around zero. There is indication of a
pattern, which suggests that the Log(X) transformation does not
fully stabilize the variance. The standardized residuals are
moderately symmetric but the presence of a large residual reduces
the normality slightly. A normal probability plot would likely show
mild deviation from the straight line. Variation in residuals tends to
increase with larger fitted values, implying heteroscedasticity,
meaning that the assumption of constant variance is not perfectly
met.
The residual analysis suggests that the curvilinear model
using Log(X) improves fit over the simple linear model, but
still exhibits:
One moderately influential observation,
Some pattern in the residuals,
Signs of increasing variance.
Therefore, the model is acceptable but not ideal, and further
model refinement (e.g., a different transformation or
weighted regression) might improve the fit.
h)
Linear regression equation:
-17.73+0.3549x300=89.17
Log model:
-152.2+91.4xlog(300)=73.91
Forecast (log model) ≈ 74 defectives
Forecast (Linear model) ≈ 89 defective
i)
Linear Model
Higher R² (0.955)
Easy to interpret
Residuals show slight curvature and increasing spread at large
batch sizes
Curvilinear Log Model
Slightly lower R² (0.921)
Residuals more normally distributed
Captures the diminishing rate of increase in defectives more
realistically
Better theoretical justification for this manufacturing context
Conclusion:
Although the linear model has a higher R², the curvilinear log
model is preferred because its residuals behave better and the
functional form aligns more closely with the practical behavior of
defect rates as batch size increases.
j) The regression model shows a strong positive relationship
between batch size and the number of defectives. As batch size
increases, defectives increase at an accelerating rate, indicating
that the process becomes less capable when handling larger
volumes. The fitted model provides good prediction accuracy, and
the residual plots show no major violations of regression
assumptions. Overall, the analysis suggests that controlling or
limiting batch size could help reduce defectives.
15.