Statistical Inference and Regression Analysis GB 3302 30
Statistical Inference and Regression Analysis: GB. 3302. 30 Professor William Greene Stern School of Business IOMS Department of Economics
Statistics and Data Analysis Part 7 – Regression Model-1 Regression Diagnostics
3/96 Using the Residuals How do you know the model is “good? ” ¢ Various diagnostics to be developed over the semester. ¢ But, the first place to look is at the residuals. ¢
4/96 Residuals Can Signal a Flawed Model Standard application: Cost function for output of a production process. ¢ Compare linear equation to a quadratic model (in logs) ¢ (123 American Electric Utilities) ¢
5/96 Electricity Cost Function
6/96 Candidate Model for Cost Log c = a + b log q + e Most of the points in this area are above the regression line. Most of the points in this area are below the regression line.
7/96 A Better Model? Log Cost = α + β 1 log. Output + β 2 [log. Output]2 + ε
8/96 Candidate Models for Cost The quadratic equation is the appropriate model. Logc = a + b 1 logq + b 2 log 2 q + e
9/96 Missing Variable Included Residuals from the quadratic cost model Residuals from the linear cost model
10/96 Unusual Data Points Outliers have (what appear to be) very large disturbances, ε Wolf weight vs. tail length The 500 most successful movies
11/96 Outliers 99. 5% of observations will lie within mean ± 3 standard deviations. We show (a+bx) ± 3 se below. ) These observations might deserve a close look. Titanic is 8. 1 standard deviations from the regression! Only 0. 86% of the 466 observations lie outside the bounds. (We will refine this later. )
12/96 Prices paid at auction for Monet paintings vs. surface area (in logs) log. Price = a + b log. Area + e Not an outlier: Monet chose to paint a small painting. Possibly an outlier: Why was the price so low?
13/96 What to Do About Outliers (1) Examine the data (2) Are they due to mismeasurement error or obvious “coding errors? ” Delete the observations. (3) Are they just unusual observations? Do nothing. (4) Generally, resist the temptation to remove outliers. Especially if the sample is large. (500 movies is large. 10 wolves is not. ) (5) Question why you think it is an outlier. Is it really?
14/96 Regression Options
15/96 Diagnostics
16/96 On Removing Outliers Be careful about singling out particular observations this way. The resulting model might be a product of your opinions, not the real relationship in the data. Removing outliers might create new outliers that were not outliers before. Statistical inferences from the model will be incorrect.
Statistics and Data Analysis Part 7 – Regression Model-2 Statistical Inference
18/96 b As a Statistical Estimator What is the interest in b? ¢ = d. E[y|x]/dx ¢ Effect of a policy variable on the expectation of a variable of interest. ¢ Effect of medication dosage on disease response ¢ … many others ¢
19/96 Application: Health Care Data German Health Care Usage Data, There altogether 27, 326 observations on German households, 1984 -1994. DOCTOR = HOSPITAL HSAT = DOCVIS = HOSPVIS = PUBLIC = ADDON = INCOME = HHKIDS = EDUC = AGE = MARRIED = EDUC = 1(Number of doctor visits > 0) = 1(Number of hospital visits > 0) health satisfaction, coded 0 (low) - 10 (high) number of doctor visits in last three months number of hospital visits in last calendar year insured in public health insurance = 1; otherwise = 0 insured by add-on insurance = 1; otherswise = 0 household nominal monthly net income in German marks / 10000. children under age 16 in the household = 1; otherwise = 0 years of schooling age in years marital status years of education
20/96 Regression? Population relationship Income = + Health + ¢ (For this population, Income =. 31237 +. 00585 Health + E[Income | Health] =. 31237 +. 00585 Health ¢
21/96 Distribution of Health
22/96 Distribution of Income
23/96 Average Income | Health Nj = 447 255 642 1173 Health 1390 4233 2570 4191 6172 3061 3192
24/96 b is a statistic Random because it is a sum of the ’s. ¢ It has a distribution, like any sample statistic ¢
25/96 Sampling Experiment 500 samples of N=52 drawn from the 27, 326 (using a random number generator to simulate N observation numbers from 1 to 27, 326) ¢ Compute b with each sample ¢ Histogram of 500 values ¢
26/96
27/96 Conclusions Sampling variability ¢ Seems to center on ¢ Appears to be normally distributed ¢
28/96 Distribution of slope estimator, b ¢ Assumptions: (Model) Regression: yi = + xi + i l (Crucial) Exogenous data: data x and noise are independent; E[ |x]=0 or Cov( , x)=0 l (Temporary) Var[ |x] = 2, not a function of x (Homoscedastic) l ¢ Results: What are the properties of b?
29/96
30/96 (1) b is unbiased and linear in
31/96 (2) b is efficient ¢ Gauss – Markov Theorem: Like Rao Blackwell. (Proof in Greene) ¢ Variance of b is smallest among linear unbiased estimators.
32/96 (3) b is consistent
33/96 Consistency: N=52 vs. N=520
34/96 a is unbiased and consistent
35/96 Covariance of a and b
36/96 Inference about Have derived expected value and variance of b. ¢ b is a ‘point’ estimator ¢ Looking for a way to form a confidence interval. ¢ Need a distribution and a pivotal statistic to use. ¢
37/96 Normality
38/96 Confidence Interval
39/96 Estimating sigma squared
40/96 Useable Confidence Interval Use s instead of s. ¢ Use t distribution instead of normal. ¢ Critical t depends on degrees of freedom ¢ b - ts < < b + ts ¢
41/96 Slope Estimator
42/96 Regression Results --------------------------------------Ordinary least squares regression. . . LHS=BOX Mean = 20. 72065 Standard deviation = 17. 49244 -----No. of observations = 62 Deg. Freedom Mean square Regression Sum of Squares = 7913. 58 1 7913. 57745 Residual Sum of Squares = 10751. 5 60 179. 19235 Total Sum of Squares = 18665. 1 61 305. 98555 -----Standard error of e = 13. 38627 Root MSE 13. 16860 Fit R-squared =. 42398 R-bar squared. 41438 Model test F[ 1, 60] = 44. 16247 Prob F > F*. 00000 ----+----------------------------------| Standard Prob. 95% Confidence BOX| Coefficient Error t |t|>T* Interval ----+----------------------------------Constant| -14. 3600** 5. 54587 -2. 59. 0121 -25. 2297 -3. 4903 CNTWAIT 3| 72. 7181*** 10. 94249 6. 65. 0000 51. 2712 94. 1650 ----+----------------------------------Note: ***, * ==> Significance at 1%, 5%, 10% level. ---------------------------------------
43/96 Hypothesis Test about Outside the confidence interval is the rejection for hypothesis tests about ¢ For the internet buzz regression, the confidence interval is ¢ 51. 2712 to 94. 1650 ¢ The hypothesis that equals zero is rejected.
Statistics and Data Analysis Part 7 -3 – Prediction
45/96 Predicting y Using the Regression Actual y 0 is + x 0 + 0 ¢ Prediction is y 0^ = a + bx 0 + 0 ¢ Error is y 0 – y 0^ = (a- ) + (b- )x 0 + 0 ¢ Variance of the error is Var[a] + x 02 Var[b] + 2 x 0 Cov[a, b] + Var[ 0] ¢
46/96 Prediction Variance
47/96 Quantum of Solace ¢ ¢ ¢ ¢ Actual Box = $67. 528882 M a=-14. 36, b=72. 7181, N=62, sb =10. 94249, s 2 = 13. 38632 buzz = 0. 76, prediction = 40. 906 Mean buzz = 0. 4824194 (buzz – mean)2 = 1. 49654 Sforecast = 13. 8314252 Confidence interval = 40. 906 +/- 2. 003(13. 831425) = 13. 239 to 68. 527 (Note: The confidence interval contains the value)
48/96 Forecasting Out of Sample Regression Analysis: G versus Income The regression equation is G = 1. 93 + 0. 000179 Income Predictor Coef SE Coef T P Constant 1. 9280 0. 1651 11. 68 0. 000 Income 0. 00017897 0. 00000934 19. 17 0. 000 S = 0. 370241 R-Sq = 88. 0% R-Sq(adj) = 87. 8% How to predict G for 2020? You would need first to predict Income for 2020. How should we do that? Per Capita Gasoline Consumption vs. Per Capita Income, 1953 -2004.
49/96 The Extrapolation Penalty The interval is narrowest at x* = , the center of our experience. The interval widens as we move away from the center of our experience to reflect the greater uncertainty. (1) Uncertainty about the prediction of x (2) Uncertainty that the linear relationship will continue to exist as we move farther from the center.
50/96 An Enduring Art Mystery Graphics show relative sizes of the two works. The Persistence of Econometrics. W. Greene, 2018 Why do larger paintings command higher prices? The Persistence of Memory. Salvador Dali, 1931
51/96 The Data Note: Using logs in this context. This is common when analyzing financial measurements (e. g. , price) and when percentage changes are more interesting than unit changes. (E. g. , what is the % premium when the painting is 10% larger? )
52/96 Monet in Large and Small Sale prices of 328 signed Monet paintings The residuals do not show any obvious patterns that seem inconsistent with the assumptions of the model. Log of $price = a + b log surface area + e
53/96 Monet Regression: There seems to be a regression. Is there a theory?
54/96 http: //www. nytimes. com/2006/05/16/arts/design/16 oran. html
55/96 167” (13 feet 11 inches) "Morning", Claude Monet 1920 -1926, oil on canvas 200 x 425 cm, Musée de l Orangerie, Paris France. Left panel 26. 2” (2 feet 2. 2”) 78. 74” (6 Feet 7 inch) 32. 1” (2 feet 8 inches)
56/96 Predicted Price for a Huge Painting
57/96 Prediction Interval for Price
58/96 118” (9 feet 10 inches) 32. 1” (2 feet 8 inches) Average Sized Monet 157” (13 Feet 1 inch) 26. 2” (2 feet 2. 2”) Use the Monet Model to Predict a Price for a Dali? Hallucinogenic Toreador
59/96
60/96 Normality Necessary for t statistics and confidence intervals ¢ Residuals reveal whether disturbances are normal? ¢ Standard tests and devices ¢
61/96 Normally Distributed Residuals? -------------------Kolmogorov-Smirnov test of F(E ) vs. Normal[. 00000, 13. 27610^2] ******* K-S test statistic =. 1810063 ******* 95% critical value =. 1727202 ******* 99% critical value =. 2070102 Normality hyp. should be rejected. -------------------
62/96 Nonnormal Disturbances ¢ Appeal to the central limit theorem ¢ Use standard normal instead of t ¢ t is essentially normal if N > 50.
Statistics and Data Analysis Part 7 -4 – Multiple Regression
64/96 Box Office and Movie Buzz 64
65/96 Box Office and Budget 65
66/96 Budget and Buzz Effects 66
67/96
68/96
69/96 An Enduring Art Mystery Graphics show relative sizes of the two works. The Persistence of Statistics. Rice, 2007 Why do larger paintings command higher prices? The Persistence of Memory. Salvador Dali, 1931
70/96 Monet Regression: There seems to be a regression. Is there a theory?
71/96 How much for the signature? ¢ The sample also contains 102 unsigned paintings Average Sale Price Signed $3, 364, 248 Not signed $1, 832, 712 ¢ Average price of signed Monet’s is almost twice that of unsigned
72/96 Can we separate the two effects? Average Prices Small Large Unsigned 346, 845 5, 795, 000 Signed 689, 422 5, 556, 490 What do the data suggest? (1) The size effect is huge (2) The signature effect is confined to the small paintings.
73/96 A Multiple Regression b 2 Ln Price = a + b 1 ln Area + b 2 (0 if unsigned, 1 if signed) + e
74/96 Monet Multiple Regression Analysis: ln (US$) versus ln (Surface. Area), Signed The regression equation is ln (US$) = 4. 12 + 1. 35 ln (Surface. Area) + 1. 26 Signed Predictor Coef SE Coef T P Constant 4. 1222 0. 5585 7. 38 0. 000 ln (Surface. Area) 1. 3458 0. 08151 16. 51 0. 000 Signed 1. 2618 0. 1249 10. 11 0. 000 S = 0. 992509 R-Sq = 46. 2% R-Sq(adj) = 46. 0% Interpretation (to be explored as we develop the topic): (1) Elasticity of price with respect to surface area is 1. 3458 – very large (2) The signature multiplies the price by exp(1. 2618) (about 3. 5), for any given size.
75/96 Ceteris Paribus in Theory ¢ Demand for gasoline: G = f(price, income) ¢ Demand (price) elasticity: e. P = %change in G given %change in P holding income constant. ¢ How do you do that in the real world? l l The “percentage changes” How to change price and hold income constant?
76/96 The Real World Data
77/96 U. S. Gasoline Market, 1953 -2004
78/96 Shouldn’t Demand Curves Slope Downward?
79/96 A Thought Experiment ¢ ¢ The main driver of gasoline consumption is income not price Income is growing over time. We are not holding income constant when we change price! How do we do that?
80/96 How to Hold Income Constant? Multiple Regression Using Price and Income Regression Analysis: G versus Gas. Price, Income The regression equation is G = 0. 134 - 0. 00163 Gas. Price + 0. 000026 Income Predictor Constant Gas. Price Income Coef 0. 13449 -0. 0016281 0. 00002634 It looks like theory works. SE Coef 0. 02081 0. 0004152 0. 00000231 T 6. 46 -3. 92 11. 43 P 0. 000
Statistics and Data Analysis Linear Multiple Regression Model
82/96 Classical Linear Regression Model ¢ The model is y = f(x 1, x 2, …, x. K, 1, 2, … K) + = a multiple regression model. Important examples: l Marginal cost in a multiple output setting l Separate age and education effects in an earnings equation. l Denote (x 1, x 2, …, x. K) as x. Boldface symbol = vector. ¢ Form of the model – E[y|x] = a linear function of x. ¢ ‘Dependent’ and ‘independent’ variables. l Independent of what? Think in terms of autonomous variation. l Can y just ‘change? ’ What ‘causes’ the change?
83/96 Model Assumptions: Generalities ¢ ¢ ¢ Linearity means linear in the parameters. We’ll return to this issue shortly. Identifiability. It is not possible in the context of the model for two different sets of parameters to produce the same value of E[y|x] for all x vectors. (It is possible for some x. ) Conditional expected value of the deviation of an observation from the conditional mean function is zero Form of the variance of the random variable around the conditional mean is specified Nature of the process by which x is observed. Assumptions about the specific probability distribution.
84/96 Linearity of the Model ¢ ¢ f(x 1, x 2, …, x. K, 1, 2, … K) = x 1 1 + x 2 2 + … + x. K K Notation: x 1 1 + x 2 2 + … + x. K K = x . l l l ¢ Boldface letter indicates a column vector. “x” denotes a variable, a function of a variable, or a function of a set of variables. There are K “variables” on the right hand side of the conditional mean “function. ” The first “variable” is usually a constant term. (Wisdom: Models should have a constant term unless theory says they should not. ) E[y|x] = 1*1 + 2*x 2 + … + K*x. K. ( 1*1 = the intercept term).
85/96 Linearity ¢ ¢ ¢ Linearity means linear in the parameters, not in the variables E[y|x] = 1 f 1(…) + 2 f 2(…) + … + K f. K(…). fk() may be any function of data. Examples: l l Logs and levels in economics Time trends, and time trends in loglinear models – rates of growth Dummy variables Quadratics, power functions, log-quadratic, trig functions, interactions and so on.
86/96 Linearity ¢ ¢ ¢ Simple linear model, E[y|x] =x’β Quadratic model: E[y|x] = α + β 1 x + β 2 x 2 Loglinear model, E[lny|x] = α + Σk lnxkβk Semilog, E[y|x] = α + Σk lnxkβk All are “linear. ” An infinite number of variations.
87/96 Matrix Notation 87
88/96 Notation Define column vectors of N observations on y and the K x variables. The assumption means that the rank of the matrix X is K. No linear dependencies => FULL COLUMN RANK of the matrix X.
89/96 An Unidentified (But Valid) Theory of Art Appreciation Enhanced Monet Area Effect Model: Height and Width Effects Log(Price) = β 1 + β 2 log Area + β 3 log Aspect Ratio + β 4 log Height + β 5 Signature + ε (Aspect Ratio = Width/Height) This X does not have full column rank. At least one column is a linear function of the others.
90/96 Conditional Homoscedasticity and Nonautocorrelation Disturbances provide no information about each other. l Var[ i | X ] = 2 l Cov[ i, j |X] = 0
91/96 Heteroscedasticity Countries are ordered by the standard deviation of their 19 residuals. Regression of log of per capita gasoline use on log of per capita income, gasoline price and number of cars per capita for 18 OECD countries for 19 years. The standard deviation varies by country.
92/96 Autocorrelation log. G=β 1 + β 2 log. Pg + β 3 log. Y + β 4 log. Pnc + β 5 log. Puc + ε
93/96 Autocorrelation Results from an Incomplete Model
94/96 Normal Distribution of ε ¢ Used to facilitate finite sample derivations of certain test statistics. ¢ Observations are independent ¢ Assumption will be unnecessary – we will use the central limit theorem for the statistical results we need.
95/96 The Linear Model ¢ y = X +ε, N observations, K columns in X, (usually including a column of ones for the intercept). l Standard assumptions about X l Standard assumptions about ε|X l E[ε|X]=0, E[ε]=0 and Cov[ε, x]=0 ¢ Regression: E[y|X] = X
Statistics and Data Analysis Least Squares
97/96 Vocabulary ¢ l l l ¢ ¢ Some terms to be used in the discussion. Population characteristics and entities vs. sample quantities and analogs Residuals and disturbances Population regression line and sample regression Objective: Learn about the conditional mean function. Estimate and 2 First step: Mechanics of fitting a line to a set of data
98/96 Least Squares
99/96 Matrix Results
100/96
101/96 Moment Matrices
102/96 Least Squares Normal Equations
103/96 --------------------------------------Ordinary least squares regression. . . LHS=G Mean = 100. 70082 Standard deviation = 14. 08790 -----No. of observations = 36 Deg. Freedom Mean square Regression Sum of Squares = 6185. 64 2 3092. 81941 Residual Sum of Squares = 760. 776 33 23. 05383 Total Sum of Squares = 6946. 42 35 198. 46901 -----Standard error of e = 4. 80144 Root MSE 4. 59703 Fit R-squared =. 89048 R-bar squared. 88384 Model test F[ 2, 33] = 134. 15642 Prob F > F*. 00000 ----+----------------------------------| Standard Prob. 95% Confidence G| Coefficient Error t |t|>T* Interval ----+----------------------------------Constant| 7. 96108 6. 23309 1. 28. 2104 -4. 72025 20. 64240 PG| -8. 86288*** 1. 35143 -6. 56. 0000 -11. 61238 -6. 11338 Y|. 01227***. 00095 12. 96. 0000. 01034. 01419 ----+---------------------------------- 103
104/96 Second Order Conditions
105/96 Does b Minimize e’e?
106/96 Positive Definite Matrix
- Slides: 106