Daily 25 - Apr 27

Class Performance

Students: 87 | Mean: 4.81 | Median: 5 | SD: 0.55

Scores ranged from 2 to 5 out of 5 points.

Correct Answer

The LASSO penalty shrinks the OLS coefficients toward zero. The purpose of \(\lambda\) is to tune the strength of the penalty.

Common Errors

“Controls” / “adjusts” for tune — Accepted.
“Push” / “drive” for shrink — Accepted as equivalent.
Reversed answers — Half credit if both concepts present but slotted in the wrong blanks.

Correct Answer

The first vertical dotted line marks the LASSO specification that minimizes MSPE (mean squared prediction error) in cross-validation.

Common Errors

“MSE” / “mean squared error” — Accepted as equivalent shorthand.
“R-squared” / “deviance” — Wrong loss function for the CV minimization shown.

Correct Answer

LASSO retained 52 coefficients (features) at the optimal \(\lambda\).

Common Errors

Correct Answer

Another year of education increases the likelihood of being in the labor force by 3.8 percentage points.

Common Errors

“3.8” without units — Half credit. Always state the units.
“3.8%” — Half credit. LPM coefficients are in probability units; “percentage points” is the precise label.
Wildly different number (e.g., 0.38, 38) — Zero credit.

Correct Answer

Logistic predicted probability = .95; LPM predicted probability = 1.04 (which is impossible, illustrating LPM’s main weakness).

Common Errors

“95%, 1.04” — Full credit. Same numerical values, just expressed as a percentage.
One of two correct — Half credit; most often students nailed .95 but missed 1.04.
Negative or wildly different values — Zero credit.

Strengths: LASSO mechanics solid | CV interpretation clear | LPM vs logistic comparison understood.

Review:

LASSO terminology — \(\lambda\) tunes; the penalty shrinks; selection is the side-effect
LPM units — Coefficients are in percentage points (probability units), not %
Why we use logistic — LPM can predict probabilities outside [0, 1], like the 1.04 case here