Regression Models - Introduction • In regression models there are two types of variables that are studied: A dependent variable, Y, also called response variable. It is modeled as random. An independent variable, X, also called predictor variable or explanatory variable. It is sometimes modeled as random and sometimes it has fixed value for each observation. • In regression models we are fitting a statistical model to data. • We generally use regression to be able to predict the value of one variable given the value of others. STA261 week 13 1 Simple Linear Regression - Introduction • Simple linear regression studies the relationship between a quantitative response variable Y, and a single explanatory variable X. • Idea of statistical model: Actual observed value of Y = … • Box (a well know statistician) claim: “All models are wrong, some are useful”. ‘Useful’ means that they describe the data well and can be used for predictions and inferences. • Recall: parameters are constants in a statistical model which we usually don’t know but will use data to estimate. STA261 week 13 2 Simple Linear Regression Models • The statistical model for simple linear regression is a straight line model of the form Y 0 1 X where… • For particular points, Yi 0 1 X i i , i 1, ..., n • We expect that different values of X will produce different mean response. In particular we have that for each value of X, the possible values of Y follow a distribution whose mean is... • Formally it means that …. STA261 week 13 3 Estimation – Least Square Method • Estimates of the unknown parameters β0 and β1 based on our observed data are usually denoted by b0 and b1. • For each observed value xi of X the fitted value of Y is yˆ i b0 b1 xi . This is an equation of a straight line. • The deviations from the line in vertical direction are the errors in prediction of Y and are called “residuals”. They are defined as ei yi yˆ i . • The estimates b0 and b1 are found by the Method of Lease Squares which is based on minimizing sum of squares of residuals. • Note, the least-squares estimates are found without making any statistical assumptions about the data. STA261 week 13 4 Derivation of Least-Squares Estimates • Let n S y i b0 b1 xi 2 i 1 • We want to find b0 and b1 that minimize S. • Use calculus…. STA261 week 13 5 Statistical Assumptions for SLR Recall, the simple linear regression model is Yi = β0 + β1Xi + εi where i = 1, …, n. The assumptions for the simple linear regression model are: 1) E(εi)=0 2) Var(εi) = σ2 3) εi’s are uncorrelated. • These assumptions are also called Gauss-Markov conditions. • The above assumptions can be stated in terms of Y’s… STA261 week 13 6 Possible Violations of Assumptions • Straight line model is inappropriate… • Var(Yi) increase with Xi…. • Linear model is not appropriate for all the data… STA261 week 13 7 Properties of Least Squares Estimates • The least-square estimates b0 and b1 are linear in Y’s. That it, there exists constants ci, di such that , b0 ciYi , b1 d iYi • Proof: Exercise.. • The least squares estimates are unbiased estimators for β0 and β1. • Proof:… STA261 week 13 8 Gauss-Markov Theorem • The least-squares estimates are BLUE (Best Linear, Unbiased Estimators). • Of all the possible linear, unbiased estimators of β0 and β1 the least squares estimates have the smallest variance. • The variance of the least-squares estimates is… STA261 week 13 9 Estimation of Error Term Variance σ2 • The variance σ2 of the error terms εi’s needs to be estimated to obtain indication of the variability of the probability distribution of Y. • Further, a variety of inferences concerning the regression function and the prediction of Y require an estimate of σ2. • Recall, for random variable Z the estimates of the mean and variance of Z based on n realization of Z are…. • Similarly, the estimate of σ2 is 1 n 2 s ei n 2 i 1 2 • S2 is called the MSE (Mean Square Error) it is an unbiased estimator of σ2. STA261 week 13 10 Normal Error Regression Model • In order to make inference we need one more assumption about εi’s. • We assume that εi’s have a Normal distribution, that is εi ~ N(0, σ2). • The Normality assumption implies that the errors εi’s are independent (since they are uncorrelated). • Under the Normality assumption of the errors, the least squares estimates of β0 and β1 are equivalent to their maximum likelihood estimators. • This results in additional nice properties of MLE’s: they are consistent, sufficient and MVUE. STA261 week 13 11 Inference about the Slope and Intercept • Recall, we have established that the least square estimates b0 and b1 are linear combinations of the Yi’s. • Further, we have showed that they are unbiased and have the following variances 1 X2 Var b0 n S XX 2 and Var b1 2 S XX • In order to make inference we assume that εi’s have a Normal distribution, that is εi ~ N(0, σ2). • This in turn means that the Yi’s are normally distributed. • Since both b0 and b1 are linear combination of the Yi’s they also have a Normal distribution. STA261 week 13 12 Inference for β1 in Normal Error Regression Model • The least square estimate of β1 is b1, because it is a linear combination of normally distributed random variables (Yi’s) we have the following result: 2 b1 ~ N 1 , S XX • We estimate the variance of b1 by S2/SXX where S2 is the MSE which has n-2 df. • Claim: The distribution b1 1 S2 of is t with n-2 df. S XX • Proof: STA261 week 13 13 Tests and CIs for β1 • The hypothesis of interest about the slope in a Normal linear regression model is H0: β1 = 0. • The test statistic for this hypothesis is b1 b1 t stat 2 S .E b1 S S XX • We compare the above test statistic to a t with n-2 df distribution to obtain the P-value…. • Further, 100(1-α)% CI for β1 is: S b1 t n 2 ; 2 b1 t n 2 ; 2 S .E b1 S XX STA261 week 13 14 Important Comment • Similar results can be obtained about the intercept in a Normal linear regression model. • However, in many cases the intercept does not have any practical meaning and therefore it is not necessary to make inference about it. STA261 week 13 15