Linear Regression Worksheet – 2007 NFL Scores/Vegas Forecasts Spread (S): Projected point differential: Away Team – Home Team (<0 Home favored) Over/Under (OU): Projected point total: Away Team + Home Team Projected Away Score: X1 = (OU+S)/2 Projected Home Score: X1 = (OU-S)/2 Other Possible Predictors: o Home Field Effect: X2=1 if Home, 0 if Away o Home Field x Forecast Interaction: X3=X1X2 Response: Y=Actual Score Regression Model: Yi = 0 + 1Xi1 + 2Xi2 + 3Xi3 + i i=1,…,512 (256 games x 2 teams) Matrix Form: Y = X + Hypothesis that the mean actual score is exactly equal to the Vegas Forecast, with no home team effects: H0: Game 1 2 3 … 254 255 256 1 2 3 … 254 255 256 i 1 2 3 … 254 255 256 257 258 259 … 510 511 512 X0 1 1 1 … 1 1 1 1 1 1 … 1 1 1 K' sse mse 49615.22 97.66775 H0: K’-m = 0 Parameter X1 28 19.25 21.75 … 17.75 26.75 20 21.5 16.75 20.75 … 25.25 21.25 14 X2 1 1 1 … 1 1 1 0 0 0 … 0 0 0 m s 9.8827 HA: K’-m ≠ 0 Estimate X3 28 19.25 21.75 … 17.75 26.75 20 0 0 0 … 0 0 0 Y 41 24 13 … 17 48 13 10 3 27 … 30 19 10 X'X 512 10696 256 5664.5 10696 232904.8 5664.5 129445.4 256 5664.5 256 5664.5 5664.5 129445.4 5664.5 129445.4 X'Y 11104 241207.8 5919 134504.5 (X'X)^-1 0.08846 -0.00430 -0.08846 0.00430 -0.00430 0.00022 0.00430 -0.00022 -0.08846 0.00430 0.21157 -0.00969 0.00430 -0.00022 -0.00969 0.00046 Beta-hat -0.377 1.050 4.453 -0.189 ^ K ' β m Q(K) 436.033 df(K) 4 K' (X' X) K 1 df(E) 508 Test Stat: _________________ Std. Error 1 t-stat P-value Rejection Region: __________ 95% Confidence Interval Lower Bound Upper Bound INTERCEPT Vegas Pick (X1) Home (X2) Vegas*Home (X3) Fitted Equation: Home _________________________________ Away _________________________ Model with only Vegas Pick (X1): X'X 512 10696 10696 232904.8 X'Y 11104 241207.75 (X'X)^-1 0.04809 -0.00221 -0.00221 0.00011 Beta-hat 1.28 0.98 Compute a 95% Confidence Interval for the mean of all games where the Vegas Pick = 25 Compute a 95% Prediction Interval for the a single (future) game where the Vegas Pick = 25 Test whether a linear relation is reasonable fit for this data (F-test for lack of Fit) # of observations = 512 df Lack of Fit Pure Error Residual # of distinct levels of X1 = 81 SS MS 9039 40682 49721 F (See spreadsheet for data) F(.05) P-value