Residual Analysis for Regression of Wages on LOS

1
Residual Analysis for Regression of Wages on LOS
No evidence that errors are not independent (or
are auto-correlated) since no trend, sine-wave,
or zig-zag alternation above below zero. Notice
3 points between 2 and 3 std. error cut-offs from
zero potential outliers but not extreme.
2
Checks for Normal Distribution of Errors
No evidence that distribution of errors is not
normal since histogram is approximately bell
shaped and normal probability plot is
approximately straight. Back up these subjective
appraisals with numerical measures. Skewness and
kurtosis both lt 1 and p-value for normal
goodness of fit gt 0.10
3
Residual Plots vs. X (LOS)
No evidence that variance of errors is not
constant since no megaphone or reverse megaphone
pattern. Also no evidence that the linear model
is the wrong functional form since no pattern
observed. Notice 3 points between 2 and 3 std.
error cut-offs from zero potential outliers but
not extreme.
4
Residual Plot vs. Yhat (predicted wages)
No evidence that variance of errors is not
constant since no megaphone or reverse megaphone
pattern. Also no evidence that the linear model
is the wrong functional form since no pattern
observed. Notice 3 points between 2 and 3 std.
error cut-offs from zero potential outliers but
not extreme.