Linear Regression Model PDF
Document Details
Uploaded by ComprehensiveChrysocolla
null
false
Damodar Gujarati
Tags
Summary
This document provides an overview of the linear regression model, including its components, assumptions, and different forms (e.g., log-linear, double log). The document is of a general nature which can be used for learning purposes. It touches on topics including sample regression and functional forms.
Full Transcript
THE LINEAR REGRESSION MODEL: AN OVERVIEW & RECAP (Gujarati, Part I, Ch. 1, or Brooks Ch. 3-4) 32 THE LINEAR REGRESSION MODEL (LPM) ➢ The general form of the LPM model is: Yi = B1 + B2X2i + B3X3i + … + BkXki + ui ➢ Or, as written in short form: Yi = BX + ui ➢ Y is the regressand, X is a vector of...
THE LINEAR REGRESSION MODEL: AN OVERVIEW & RECAP (Gujarati, Part I, Ch. 1, or Brooks Ch. 3-4) 32 THE LINEAR REGRESSION MODEL (LPM) ➢ The general form of the LPM model is: Yi = B1 + B2X2i + B3X3i + … + BkXki + ui ➢ Or, as written in short form: Yi = BX + ui ➢ Y is the regressand, X is a vector of regressors, and u is an error term. ➢ The subscript i denotes the ith observation Damodar Gujarati Econometrics by Example, second edition 33 On the Meaning of Linear Regression • Refers to linearity in the regression coefficients, the Bs, and not linearity in the Y and X variables • For instance, the Y and X variables can be logarithmic (e.g. ln X2), or reciprocal (1/X3) or raised to a power (e.g. X2) • Linearity in the B coefficients means that they are not raised to any power (e.g. 𝐵12 ) or divided by other coefficients (e.g. B2/B3) or transformed, such as ln B4 • There are occasions where regression models that are not linear in the regression coefficients are considered (non-linear econometrics; out of the scope of this course) 34 POPULATION (TRUE) MODEL Yi = B1 + B2X2i + B3X3i + … + BkXki + ui ➢ This equation is known as the population or true model. ➢ It consists of two components: ➢ (1) A deterministic component, BX (the conditional mean of Y, or E(Y|X)). ➢ (2) A nonsystematic, or random component, ui. Damodar Gujarati Econometrics by Example, second edition 35 REGRESSION COEFFICIENTS ➢ B1 is the intercept. ➢ B2 to Bk are the slope coefficients. ➢ Collectively, they are the regression coefficients or regression parameters. ➢ Each slope coefficient measures the (partial) rate of change in the mean value of Y for a unit change in the value of a regressor, ceteris paribus. ➢ Any causal relationship between Y and the Xs, should be based on the relevant theory Damodar Gujarati Econometrics by Example, second edition 36 SAMPLE REGRESSION FUNCTION ➢ The sample counterpart is: Yi = b1 + b2X2i + b3X3i + … + bkXki + ei ➢ Or, as written in short form: Yi = bX + ei where e is a residual. ➢ The deterministic component is written as: Yi = b1 + b2 X 2i + b3 X 3i + ... + bk X ki = bX ➢ 𝑌𝑖 = 𝑌𝑖 + 𝑒𝑖 Estimated value = Determistic value + residual Damodar Gujarati Econometrics by Example, second edition 37 Terminological issues • Alternative way to put it: • Regressand = dependent variable (= response variable) • Regressor = independent variable = explanatory variable • b1 (or B1) is often denoted as b0 (or Bo) • The deterministic component can be thought of as the expected value of Y given X [i.e. E(Y|X)] • b coefficients are the estimators of the (true) B coefficients 38 THE NATURE OF DATA ➢ Cross-Section Data ➢ Data on one or more variables collected at the same point in time. ➢ Examples are the census of population conducted by the Census Bureau every 10 years, opinion polls conducted by various polling organizations, and temperature at a given time in several places. Damodar Gujarati Econometrics by Example, second edition 39 THE NATURE OF DATA ➢ Time Series Data ➢ A set of observations that a variable takes at different times, such as daily (e.g., stock prices), weekly (e.g., money supply), monthly (e.g., the unemployment rate), quarterly (e.g., GDP), annually (e.g., government budgets), quinquenially or every five years (e.g., the census of manufactures), or decennially or every ten years (e.g., the census of population). Damodar Gujarati Econometrics by Example, second edition 40 THE NATURE OF DATA ➢ Panel, Longitudinal or Micro-panel Data ➢ Combines features of both cross-section and time series data. ➢ Same cross-sectional units are followed over time. ➢ Panel data represents a special type of pooled data (simply time series, cross-sectional, where the same cross-sectional units are not necessarily followed over time). Damodar Gujarati Econometrics by Example, second edition 41 Macro Data • Macro data: data at the national level • E.g. GDP, inflation rate, etc. • Often refers to market level data; e.g. housing price index for Oulu, OMX Helsinki index returns, etc. • Macro econometrics • Tools and techniques needed to model aggregate economic data, e.g. unemployment, wages, prices • In practice, typically based on the methods of time series econometrics 42 Micro Data • Micro data: data on the characteristics of units of a population, such as individuals, households, or establishments • E.g. income at the household level, data on individual housing transactions • (Was) Conventionally largely correlation based – causality complications • More recently, approaches aiming to identify causal relations: diff-in-diff, Regression discontinuity design (RDD) • Economics Nobel prize for Joshua D. Angrist, David Card, and Guido W. Imbens in 2021 (https://www.nber.org/news/joshua-angrist-david-cardand-guido-imbens-awarded-2021-nobel-prize ) 43 METHOD OF ORDINARY LEAST SQUARES ➢ Method of Ordinary Least Squares (OLS) does not minimize the sum of the error term, but minimizes error sum of squares (ESS): 2 2 u = ( Y − B − B X − B X − .... − B X ) i i 1 2 2 i 3 3i k ki ➢ To obtain values of the regression coefficients, derivatives are taken with respect to the regression coefficients and set equal to zero. Damodar Gujarati Econometrics by Example, second edition 44 CLASSICAL LINEAR REGRESSION MODEL ➢ Assumptions of the Classical Linear Regression Model (CLRM): ➢ A-1: Model is linear in the parameters. ➢ A-2: Regressors (RHS variables) are fixed or nonstochastic.* ➢ A-3: Given X, the expected value of the error term is zero, or E(ui |X) = 0. * In the sense that their values are fixed in repeated sampling Damodar Gujarati Econometrics by Example, second edition 45 CLASSICAL LINEAR REGRESSION MODEL ➢ A-4: Homoscedastic, or constant, variance of ui, or var(ui|X) = σ2. ➢ A-5: No autocorrelation, or cov(ui,uj|X) = 0, i ≠ j. ➢ A-6: No multicollinearity, or no perfect linear relationships among the X variables. ➢ A-7: No specification bias. Damodar Gujarati Econometrics by Example, second edition 46 GAUSS-MARKOV THEOREM ➢ On the basis of assumptions A-1 to A-7, the OLS method gives best linear unbiased estimators (BLUE): ➢ (1) Estimators are linear functions of the dependent variable Y. ➢ (2) The estimators are unbiased; in repeated applications of the method, the estimators approach their true values. ➢ (3) In the class of linear estimators, OLS estimators have minimum variance; i.e., they are efficient, or the “best” estimators. (→ the true parameter values can be estimated with least possible uncertainty; an unbiased estimator with the least variance is called an efficient estimator) ➢ For more details, see Gujarati 1.4 (pp. 8-10). Damodar Gujarati Econometrics by Example, second edition 47 Violations of the assumptions ➢ Often one or more of the assumptions are violated. ➢ This does not mean that OLS could not be used at all. ➢ In many cases, there are techniques to cater for the violations - or at least to diminish their biasing influence. ➢ These will be discussed especially in lecture (handout) II. ➢ Also considered within many of the topics later during the course. Damodar Gujarati Econometrics by Example, second edition 48 Hypothesis testing • We do not know the true model – instead, the aim is to estimate a model that reflects the true one • That is, the estimated parameters are only estimates (estimators of the true coefficients), not accurate true values • OLS estimators, bs, are random variables, for their values will vary from sample to sample • Hence, we need statistical tests and confidence intervals in order to make conclusions / to derive implications 49 HYPOTHESIS TESTING: t TEST ➢ To test the following hypothesis: H0: Bk = 0 H1: Bk ≠ 0 we calculate the following and use the t table to obtain the critical t value with n-k degrees of freedom for a given level of significance (or α, equal to 10%, 5%, or 1%): bk t= se(bk ) If this value is greater than the critical t value, we can reject H0. On the estimation of se(bk), Gujarati 1.5 (p. 10) Damodar Gujarati Econometrics by Example, second edition 50 HYPOTHESIS TESTING: t TEST ➢ An alternative method is seeing whether zero lies within the confidence interval: [bk t / 2 se(bk )] = (1 − ) ➢ If zero lies in this interval, we cannot reject H0. ➢ The p-value gives the exact level of significance, or the lowest level of significance at which we can reject H0. Damodar Gujarati Econometrics by Example, second edition 51 Type I and II Errors • Type I error is the rejection of a true null hypothesis (also known as a "false positive" finding or conclusion) • Type II error is the non-rejection of a false null hypothesis (also known as a "false negative" finding or conclusion) • The size of a test is the probability of committing a Type I error, i.e., of incorrectly rejecting the null hypothesis when the null hypothesis is true • The power of a binary hypothesis test is the probability of correctly rejecting the null hypothesis if it is false — i.e., it indicates the probability of avoiding a type II error 52 GOODNESS OF FIT, R2 ➢ R2, the coefficient of determination, is an overall measure of goodness of fit of the estimated regression line. ➢ Gives the percentage of the total variation in the dependent variable that is explained by the regressors. ➢ It is a value between 0 (no fit) and 1 (perfect fit). ➢ Let: Explained Sum of Squares (ESS) = (Yˆ − Y ) 2 Residual Sum of Squares (RSS) = e 2 Total Sum of Squares (TSS) = (Y − Y ) 2 ➢ Then: ESS RSS R = = 1− TSS TSS 2 Damodar Gujarati Econometrics by Example, second edition 53 HYPOTHESIS TESTING: F TEST ➢ Testing the following hypothesis is equivalent to testing the hypothesis that all the slope coefficients are 0: H0: R2 = 0 H1: R2 ≠ 0 ➢ Calculate the following and use the F table to obtain the critical F value with k-1 degrees of freedom in the numerator and n-k degrees of freedom in the denominator for a given level of significance: ESS / df R 2 /(k − 1) F= = RSS / df (1 − R 2 ) /(n − k ) If this value is greater than the critical F value, reject H0. n = number of usable observations; k = number of regressors Damodar Gujarati Econometrics by Example, second edition 54 On the t and F tests • In sum, in the basic linear regression model • t-test concerns the hypothesis that a given single regression coefficient has a specific value (often, but not necessarily zero) • F-test concerns the whole regression: whether the regression model has any explanatory power at all w.r.t. to Y 55 Functional Forms of Regression Models (Gujarati, Part I, Ch 2) LOG-LINEAR, DOUBLE LOG, OR CONSTANT ELASTICITY MODELS ➢ The Cobb-Douglas Production Function: Qi = B1Li B2 Ki B3 can be transformed into a linear model by taking natural logs of both sides: ln Q = ln B + B ln L + B ln K i 1 2 i 3 i ➢ The slope coefficients can be interpreted as elasticities. ➢ If (B2 + B3) = 1, we have constant returns to scale. ➢ If (B2 + B3) > 1, we have increasing returns to scale. ➢ If (B2 + B3) < 1, we have decreasing returns to scale. Damodar Gujarati Econometrics by Example, second edition 57 LOG-LIN MODELS ➢ Log-lin models follow this general form: ln 𝑌 = 𝐵1 + 𝐵2 𝑋𝑖 ➢ B2 is the relative (or percentage) change in Y responding to a one unit absolute change in X ➢ If X increases by 1, predicted Y increases by B2 % ➢ Used in Engel expenditure functions: “The total expenditure that is devoted to food tends to increase in arithmetic progression as total expenditure increases in geometric proportion.” Damodar Gujarati Econometrics by Example, second edition 58 LIN-LOG MODELS ➢ Lin-log models follow this general form: Yi = B1 + B2 ln X i + ui ➢ B2 is the absolute change in Y responding to a 100% change in X ➢ If X increases by 100%, predicted Y increases by B2 units Damodar Gujarati Econometrics by Example, second edition 59 RECIPROCAL MODELS ➢ Sometimes the relationship between the regressand and regressor(s) is reciprocal or inverse: 1 Yi = B1 + B2 ( ) + ui Xi ➢ Note that: 1 ➢ As X increases indefinitely, the term B2 ( X i ) approaches zero and Y approaches the limiting or asymptotic value B1. 1 ➢ The slope is: dY = − B2 ( 2 ) dX X ➢ Therefore, if B2 is positive, the slope is negative throughout, and if B2 is negative, the slope is positive throughout. Damodar Gujarati Econometrics by Example, second edition 60 POLYNOMIAL REGRESSION MODELS ➢ The following regression predicting housing prices (HP) is an example of a quadratic function, or more generally, a seconddegree polynomial in the variable size: 𝐻𝑃 = 𝐵0 + 𝐵1 𝑠𝑖𝑧𝑒 + 𝐵2 𝑠𝑖𝑧𝑒 2 ➢ The slope is nonlinear and equal to: 𝑑𝐻𝑃 = 𝐵1 + 2𝐵2 𝑠𝑖𝑧𝑒 𝑠𝑖𝑧𝑒 61 SUMMARY OF FUNCTIONAL FORMS MODEL FORM SLOPE ELASTICITY dY ) dX dY X . dX Y B2 B2 ( ( X ) Y Linear Y =B1 + B2 X Log-linear lnY =B1 + ln X B2 ( Log-lin lnY =B1 + B2 X B2 (Y ) B2 ( X ) Lin-log Y = B1 + B2 ln X B2 ( 1 ) X 1 B2 ( ) Y 1 ) X − B2 ( 1 ) 2 X Reciprocal Y = B1 + B2 ( Y ) X B2 − B2 ( 1 ) XY NOTE: We cannot directly compare the fits of two models that have different dependent variables, but we can transform the models and compare RSS (see Gujarati, 2.8) Damodar Gujarati Econometrics by Example, second edition 62 STANDARDIZED VARIABLES ➢ We can avoid the problem of having variables measured in different units by expressing them in standardized form: _ − Yi − Y Xi − X * * Yi = ; Xi = SY SX _ _ where SY and SX are the sample standard deviations and Y and X are the sample means of Y and X, respectively ➢ The mean value of a standardized variable is always zero and its standard deviation value is always 1. ➢ Gujarati pp. 41-43 Damodar Gujarati Econometrics by Example, second edition 63 MEASURES OF GOODNESS OF FIT ➢ R2: Measures the proportion of the variation in the regressand explained by the regressors.− ➢ Adjusted R2: Denoted as R 2, it takes degrees of freedom into account: _ 2 2 n −1 POLL R = 1 − (1 − R ) n−k ➢ Various Information Criteria ➢ Akaike’s Information Criterion (AIC) ➢ Schwarz’s Information Criterion (SIC) ➢ Also other info criteria ➢ To be discussed more in the time series econometrics “stage” 64 Qualitative Explanatory Variables in Regression Models (Gujarati, Part I, Ch 3) QUALITATIVE VARIABLES ➢ Qualitative variables are nominal scale variables which have no particular numerical values. ➢ We can “quantify” them by creating the so-called dummy variables, which take values of 0 and 1 ➢ 0 indicates the absence of an attribute ➢ 1 indicates the presence of the attribute ➢ For example, a variable denoting gender can be quantified as female = 1 and male = 0 or vice versa. ➢ Dummy variables are also called indicator variables, categorical variables, and qualitative variables. ➢ Examples: gender, race, color, religion, nationality, geographical region, party affiliation, and political upheavals Damodar Gujarati Econometrics by Example, second edition 66 DUMMY VARIABLE TRAP ➢ If an intercept is included in the model and if a qualitative variable has m categories, then introduce only (m – 1) dummy variables. ➢ For example, gender has only two categories; hence we introduce only one dummy variable for gender. ➢ This is because if a female gets a value of 1, ipso facto a male gets a value of zero. ➢ If we consider self-reported health as a choice among excellent, good, and poor, we can have at most two dummy variables to represent the three categories. ➢ If we do not follow this rule, we will fall into what is called the dummy variable trap, the situation of perfect collinearity. Damodar Gujarati Econometrics by Example, second edition 67 REFERENCE CATEGORY ➢ The category that gets the value of 0 is called the reference, benchmark, or comparison category. ➢ All comparisons are made in relation to the reference category. ➢ If there are several dummy variables, you must keep track of the reference category; otherwise, it will be difficult to interpret the results. Damodar Gujarati Econometrics by Example, second edition 68 POINTS TO KEEP IN MIND ➢ If there is an intercept in the regression model, the number of dummy variables must be one less than the number of classifications of each qualitative variable. ➢ If you drop the (common) intercept from the model, you can have as many dummy variables as the number of categories of the dummy variable. ➢ The coefficient of a dummy variable must always be interpreted in relation to the reference (i.e. omitted) category. ➢ Dummy variables can interact with quantitative regressors as well as with qualitative regressors. If a model has several qualitative variables with several categories, introduction of dummies for all the combinations can consume a large number of degrees of freedom. Damodar Gujarati Econometrics by Example, second edition 69 INTERPRETATION OF DUMMY VARIABLES ➢ Dummy coefficients are often called differential intercept dummies, for they show the differences in the intercept values of the category that gets the value of 1 as compared to the reference category. ➢ The common intercept value refers to all those categories that take a value of 0. Damodar Gujarati Econometrics by Example, second edition 70 USE OF DUMMY VARIABLES IN SEASONAL DATA ➢ The process of removing the seasonal component from a time series is called deseasonalization or seasonal adjustment. ➢ The resulting time series is called deseasonalized or seasonally adjusted time series. ➢ Consider the following model predicting the sales of fashion clothing: Sales = A + A D + A D + A D + u t 1 2 2t 3 3t 4 4t t where D2 =1 for second quarter, D3 =1 for third quarter, D4= 1 for 4th quarter, Sales = real sales per thousand square feet of retail space. Damodar Gujarati Econometrics by Example, second edition 71 USE OF DUMMY VARIABLES IN SEASONAL DATA (pp. 58-61) ➢ In order to deseasonalize the sales time series, we proceed as follows: ➢ 1. From the estimated model we obtain the estimated sales volume. ➢ 2. Subtract the estimated sales value from the actual sales volume and obtain the residuals. ➢ 3. To the estimated residuals, we add the (sample) mean value of sales. The resulting values are the deseasonalized sales values. Damodar Gujarati Econometrics by Example, second edition 72 ➢ Dummy variables also can be used e.g. to capture structural changes in data or parameters over time Damodar Gujarati Econometrics by Example, second edition 73