Advanced Regression Summer Statistics Institute. Day 2: MLR and Dummy Variables

Transcription

1 Advanced Regression Summer Statistics Institute Day 2: MLR and Dummy Variables 1

2 The Multiple Regression Model Many problems involve more than one independent variable or factor which affects the dependent or response variable. More than size to predict house price! Demand for a product given prices of competing brands, advertising,house hold attributes, etc. In SLR, the conditional mean of Y depends on X. The Multiple Linear Regression (MLR) model extends this idea to include more than one independent variable. 2

3 The MLR Model Same as always, but with more covariates. Y = β 0 + β 1 X 1 + β 2 X β p X p + ɛ Recall the key assumptions of our linear regression model: (i) The conditional mean of Y is linear in the X j variables. (ii) The error term (deviations from line) are normally distributed independent from each other identically distributed (i.e., they have constant variance) Y X 1... X p N(β 0 + β 1 X β p X p, σ 2 ) 3

4 The MLR Model Our interpretation of regression coefficients can be extended from the simple single covariate regression case: β j = E[Y X 1,..., X p ] X j Holding all other variables constant, β j is the average change in Y per unit change in X j. 4

5 The MLR Model If p = 2, we can plot the regression surface in 3D. Consider sales of a product as predicted by price of this product (P1) and the price of a competing product (P2). Sales = β 0 + β 1 P1 + β 2 P2 + ɛ 5

6 Least Squares Y = β 0 + β 1 X β p X p + ε, ε N(0, σ 2 ) How do we estimate the MLR model parameters? The principle of Least Squares is exactly the same as before: Define the fitted values Find the best fitting plane by minimizing the sum of squared residuals. 6

7 Least Squares The data... p1 p2 Sales

8 Least Squares Model: SUMMARY OUTPUT Sales i = β 0 + β 1 P1 i + β 2 P2 i + ɛ i, ɛ N(0, σ 2 ) Regression Statistics Multiple R 0.99 R Square 0.99 Adjusted R Square 0.99 Standard Error Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept p p b 0 = ˆβ 0 = , b 1 = ˆβ 1 = 97.66, b 2 = ˆβ 2 = , s = ˆσ =

9 Plug-in Prediction in MLR Suppose that by using advanced corporate espionage tactics, I discover that my competitor will charge $10 the next quarter. After some marketing analysis I decided to charge $8. How much will I sell? Our model is Sales = β 0 + β 1 P1 + β 2 P2 + ɛ with ɛ N(0, σ 2 ) Our estimates are b 0 = 115, b 1 = 97, b 2 = 109 and s = 28 which leads to Sales = P P2 + ɛ with ɛ N(0, 28 2 ) 9

10 Plug-in Prediction in MLR By plugging-in the numbers, Sales = ɛ = ɛ Sales P1 = 8, P2 = 10 N(437, 28 2 ) and the 95% Prediction Interval is (437 ± 2 28) 381 < Sales <

11 Least Squares Just as before, each b i is our estimate of β i Fitted Values: Ŷ i = b 0 + b 1 X 1i + b 2 X 2i... + b p X p. Residuals: e i = Y i Ŷi. Least Squares: Find b 0, b 1, b 2,..., b p to minimize n i=1 e2 i. In MLR the formulas for the b i s are too complicated so we won t talk about them... 11

12 Least Squares 12

13 Least Squares p1 p2 Sales yhat residuals b0 b1 b Excel break: fitted values, residuals,... 13

14 Residual Standard Error The calculation for s 2 is exactly the same: s 2 = n i=1 e2 i n p 1 = n i=1 (Y i Ŷi) 2 n p 1 Ŷ i = b 0 + b 1 X 1i + + b p X pi The residual standard error is the estimate for the standard deviation of ɛ,i.e, ˆσ = s = s 2. 14

15 Residuals in MLR As in the SLR model, the residuals in multiple regression are purged of any linear relationship to the independent variables. Once again, they are on average zero. Because the fitted values are an exact linear combination of the X s they are not correlated to the residuals. We decompose Y into the part predicted by X and the part due to idiosyncratic error. Y = Ŷ + e ē = 0; corr(x j, e) = 0; corr(ŷ, e) = 0 15

16 Residuals in MLR Consider the residuals from the Sales data: residuals residuals residuals fitted P P2 16

17 Fitted Values in MLR Another great plot for MLR problems is to look at Y (true values) against Ŷ (fitted values). y=sales y.hat (MLR: p1 and p2) If things are working, these values should form a nice straight line. Can you guess the slope of the blue line? 17

18 Fitted Values in MLR With just P1... y=sales y=sales y=sales p y.hat (SLR: p1) 0 Left plot: Sales vs P1 Right plot: Sales vs. ŷ (only P1 as a regressor) 18

19 Fitted Values in MLR Now, with P1 and P2... y=sales y=sales y=sales y.hat(slr:p1) y.hat(slr:p2) y.hat(mlr:p1 and p2) First plot: Sales regressed on P1 alone... Second plot: Sales regressed on P2 alone... Third plot: Sales regressed on P1 and P2 19

20 R-squared We still have our old variance decomposition identity... SST = SSR + SSE... and R 2 is once again defined as R 2 = SSR SST = 1 SSE SST telling us the percentage of variation in Y explained by the X s. In Excel, R 2 is in the same place and Multiple R refers to the correlation between Ŷ and Y. 20

21 Least Squares Sales SUMMARY i = OUTPUT β 0 + β 1 P1 i + β 2 P2 i + ɛ i Regression Statistics Multiple R 0.99 R Square 0.99 Adjusted R Square 0.99 Standard Error Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept p p R 2 = 0.99 Multiple R = r Y,Ŷ = corr(y,ŷ ) = 0.99 Note that R 2 = corr(y, Ŷ ) 2 21

22 Back to Baseball SUMMARY OUTPUT R/G = β 0 + β 1 OBP + β 2 SLG + ɛ Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 30 ANOVA df SS MS F Significance F Regression E 15 Residual Total Coefficientst andard Erro t Stat P value Lower 95% Upper 95% Intercept E OBP E SLG R 2 = Multiple R = r Y,Ŷ = corr(y,ŷ ) = Note that R 2 = corr(y, Ŷ ) 2 22

23 Intervals for Individual Coefficients As in SLR, the sampling distribution tells us how close we can expect b j to be from β j The LS estimators are unbiased: E[b j ] = β j for j = 0,..., d. We denote the sampling distribution of each estimator as b j N(β j, s 2 b j ) 23

24 Intervals for Individual Coefficients Intervals and t-statistics are exactly the same as in SLR. A 95% C.I. for β j is approximately b j ± 2s bj The t-stat: t j = (b j β 0 j ) s bj is the number of standard errors between the LS estimate and the null value (β 0 j ) As before, we reject the null when t-stat is greater than 2 in absolute value Also as before, a small p-value leads to a rejection of the null Rejecting when the p-value is less than 0.05 is equivalent to rejecting when the t j > 2 24

25 Intervals for Individual Coefficients IMPORTANT: Intervals and testing via b j & s bj procedures: are one-at-a-time You are evaluating the j th coefficient conditional on the other X s being in the model, but regardless of the values you ve estimated for the other b s. 25

26 In Excel... Do we know all of these numbers? SUMMARY OUTPUT Regression Statistics Multiple R 0.99 R Square 0.99 Adjusted R Square 0.99 Standard Error Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept p p % C.I. for β 1 b1 ± 2 s b1 [ ; ] = [ ; 92.36] 26

27 Understanding Multiple Regression There are two, very important things we need to understand about the MLR model: 1. How dependencies between the X s affect our interpretation of a multiple regression; 2. How dependencies between the X s inflate standard errors (aka multicolinearity) We will look at a few examples to illustrate the ideas... 27

28 Understanding Multiple Regression The Sales Data: Sales : units sold in excess of a baseline P1: our price in $ (in excess of a baseline price) P2: competitors price (again, over a baseline) 28

29 Understanding Multiple Regression ress If we regress Sales on our own price, we obtain a somewhat Regression Plot surprising conclusion... thesales higher = the price p1 the more we sell!! S = R-Sq = 19.6 % R-Sq(adj) = 18.8 % e, in the at g on gher ssociated re sales!! Sales It looks like we should just raise our prices, right? NO, not if 4 p1 you have taken this statistics class! 29 The regression line

30 Understanding Multiple Regression The regression equation for Sales on own price (P1) is: Sales = P1 If now we add the competitors price to the regression we get Sales = P P2 Does this look better? How did it happen? Remember: 97.7 is the affect on sales of a change in P1 with P2 held fixed!! 30

31 Understanding Multiple Regression How can we see what is going on? How can we see what is going on? Let s compare Sales in two If we different compares observations: sales in weeks weeks and and , we see that an increase in p1, holding p2 constant We see that an increase in P1, holding P2 constant, (82 to 99) corresponds to a drop is sales. corresponds to a drop in Sales! p p Sales p1 Note the strong relationship between p1 and p2!! Note the strong relationship (dependence) between P1 and P2!!

32 Understanding Multiple Regression Here we select a subset of points where p varies and Let s p2 does look atis ahelp subsetapproximately of points where P1constant. varies and P2 is held approximately constant p1 5 4 Sales p p For a fixed level of p2, variation in p1 is negatively correlated with Sales!! with sale! For a fixed level of P2, variation in P1 is negatively correlated 32

33 Understanding Multiple Regression Different colors indicate different ranges of p2. Below, different colors indicate different ranges for P2... p1 larger p1 are associated with larger p2 Sales for each fixed level of p2 there is a negative relationship between sales and p1 sales$p sales$sales sales$p2 p sales$p1 p1 33

34 Understanding Multiple Regression Summary: 1. A larger P1 is associated with larger P2 and the overall effect leads to bigger sales 2. With P2 held fixed, a larger P1 leads to lower sales 3. MLR does the trick and unveils the correct economic relationship between Sales and prices! 34

35 Understanding Multiple Regression Beer Data (from an MBA class) nbeer number of beers before getting drunk d height and weight 20 nbeer height Is number of beers related to height? 35

36 Understanding Multiple Regression SUMMARY OUTPUT nbeers = β 0 + β 1 height + ɛ Regression Statistics Multiple R 0.58 R Square 0.34 Adjusted R Square 0.33 Standard Error 3.11 Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept height Yes! Beers and height are related... 36

37 Understanding Multiple Regression nbeers = β 0 + β 1 weight + β 2 height + ɛ SUMMARY OUTPUT Regression Statistics Multiple R 0.69 R Square 0.48 Adjusted R Square 0.46 Standard Error 2.78 Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept weight height What about now?? Height is not necessarily a factor... 37

38 S = R-Sq = 48.1% R-Sq(adj) = 45.9% Understanding Multiple Regression 75 The correlations: height nbeer weight weight height weight 200 The two x s are highly correlated!! If we regress beers only on height we see an effect. Bigger heights go with more beers. However, when height goes up weight tends to go up as well... in the first regression, height was a proxy for the real cause of drinking ability. Bigger people can drink more and weight is a more accurate measure of bigness. 38

39 No, not all. Understanding Multiple Regression weight S = R-Sq = 48.1% R-Sq(adj) = 45.9% 75 The correlations: height nbeer weight weight height weight 200 The two x s are highly correlated!! In the multiple regression, when we consider only the variation in height that is not associated with variation in weight, we see no relationship between height and beers. 39

40 Understanding Multiple Regression SUMMARY OUTPUT nbeers = β 0 + β 1 weight + ɛ Regression Statistics Multiple R 0.69 R Square 0.48 Adjusted R 0.47 Standard E 2.76 Observation 50 ANOVA df SS MS F Significance F Regression E-08 Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept weight Why is this a better model than the one with weight and height?? 40

41 Understanding Multiple Regression In general, when we see a relationship between y and x (or x s), that relationship may be driven by variables lurking in the background which are related to your current x s. This makes it hard to reliably find causal relationships. Any correlation (association) you find could be caused by other variables in the background... correlation is NOT causation Any time a report says two variables are related and there s a suggestion of a causal relationship, ask yourself whether or not other variables might be the real reason for the effect. Multiple regression allows us to control for all important variables by including them into the regression. Once we control for weight, height and beers are NOT related!! 41

42 Understanding Multiple Regression With the above examples we saw how the relationship amongst the X s can affect our interpretation of a multiple regression... we will now look at how these dependencies will inflate the standard errors for the regression coefficients, and hence our uncertainty about them. Remember that in simple linear regression our uncertainty about b 1 is measured by s 2 b 1 = s 2 (n 1)s 2 x = s 2 n i=1 (X i X ) 2 The more variation in X (the larger s 2 x ) the more we know about β 1... ie, (b 1 β 1 ) is smaller. 42

43 Understanding Multiple Regression In Multiple Regression we seek to relate the variation in Y to the variation in an X holding the other X s fixed. So, we need to see how much each X varies on its own. in MLR, the standard errors are defined by the following formula: s 2 b j = s 2 variation in X j not associated with other X s How do we measure the bottom part of the equation? We regress X j on all the other X s and compute the residual sum of squares (call it SSE j ) so that s 2 b j = s2 SSE j 43

44 Understanding Multiple Regression In the number of beers example... s = 2.78 and the regression on height on weight gives... SUMMARY OUTPUT Regression Statistics Multiple R R Square Adjusted R Squar Standard Error Observations ANOVA df SS MS F Significance F Regression Residual Total Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept weight SSE 2 = s b2 = = 0.20 Is this right? 44

45 Understanding Multiple Regression What happens if we are regressing Y on X s that are highly correlated. SSE j goes down and the standard error s bj up! goes What is the effect on the confidence intervals (b j ± 2 s bj )? They get wider! This situation is called Multicolinearity If a variable X does nothing on its own we can t estimate its effect on Y. 45

46 SUMMARY OUTPUT Back to Baseball Let s try to add AVG on top of OBP Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 30 ANOVA df SS MS F Significance F Regression E 14 Residual Total Coefficientst andard Erro t Stat P value Lower 95% Upper 95% Intercept E AVG OBP E R/G = β 0 + β 1 AVG + β 2 OBP + ɛ Is AVG any good? 46

47 SUMMARY OUTPUT Back to Baseball - Now let s add SLG Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 30 ANOVA df SS MS F Significance F Regression E 15 Residual Total Coefficientst andard Erro t Stat P value Lower 95% Upper 95% Intercept E OBP E SLG R/G = β 0 + β 1 OBP + β 2 SLG + ɛ What about now? Is SLG any good 47

48 Back to Baseball Correlations AVG 1 OBP SLG When AVG is added to the model with OBP, no additional information is conveyed. AVG does nothing on its own to help predict Runs per Game... SLG however, measures something that OBP doesn t (power!) and by doing something on its own it is relevant to help predict Runs per Game. (Okay, but not much...) 48

49 F-tests In many situation, we need a testing procedure that can address simultaneous hypotheses about more than one coefficient Why not the t-test? We will look at two important types of simultaneous tests (i) Overall Test of Significance (ii) Partial F-test The first test will help us determine whether or not our regression is worth anything... the second will allow us to compare different models. 49

50 Supervisor Performance Data Suppose you are interested in the relationship between the overall performance of supervisors to specific activities involving interactions between supervisors and employees (from a psychology management study) The Data Y = Overall rating of supervisor X 1 = Handles employee complaints X 2 = Does not allow special privileges X 3 = Opportunity to learn new things X 4 = Raises based on performance X 5 = Too critical of poor performance X 6 = Rate of advancing to better jobs 50

51 Supervisor Performance Data Supervisor Performance Data Applied Regression Analysis Carlos M. Carvalho!""#$%&'$()*+"$,- 51

52 Supervisor Performance Data F-tests %.$/0"1"$234$1")2/*53.0*6$0"1"7$81"$2))$/0"$95"::*9*"3/.$.*;3*:*923/7$!02/$2<5=/$/0"$;15=6$5:$>21*2<)".$953.*+"1"+$/5;"/0"17 Is there any relationship here? Are all the coefficients significant? Applied Regression Analysis Carlos M. Carvalho What about all of them together?!""#$%&'$()*+"$,- 52

53 Why not look at R 2 R 2 in MLR is still a measure of goodness of fit. However it ALWAYS grows as we increase the number of explanatory variables. Even if there is no relationship between the X s and Y, R 2 > 0!! To see this let s look at some Garbage Data 53

54 F-tests Garbage Data./$0""$12*03$)"140$5"6"781"$0/9"$587:85"$+818$1281$280$6/12*65$1/$ I made up 6 garbage variables that have nothing to do with Y... D*701$)"140$98#"$E=$0/9"$F587:85"G$@HI3$H,3$J3$HKB$<87*8:)"0C Applied Regression Analysis Carlos M. Carvalho!""#$%&'$()*+"$,- 54

55 Garbage Data R 2 is 26%!! We need to develop a way to see whether a R 2 of 26% can happen by chance when all the true β s are zero. It turns out that if we transform R 2 we can solve this. Define f = R 2 /p (1 R 2 )/(n p 1) A big f corresponds to a big R 2 but there is a distribution that tells what kind of f we are likely to get when all the coefficients are indeed zero... The f statistic provides a scale that allows us to decide if big is big enough. 55

56 The F -test We are testing: H 0 : β 1 = β 2 =... β p = 0 H 1 : at least one β j 0. This is the F-test of overall significance. Under the null hypothesis f is distributed: f F p,n p 1 Generally, f > 4 is very significant (reject the null). 56

57 The F -test What kind of distribution is this? F dist. with 6 and 23 df density It is a right skewed, positive valued family of distributions indexed by two parameters (the two df values). 57

58 The F-test # 5 ((7 6 # 5 Two equivalent expressions for f 9"3:;$<="<#$3=*;$3";3$4/>$3="$?@A>BA@"C$DA>*AB)";E Let s check this test for the garbage data... F-tests./0$12/34$45"$/6*7*81)$181)9:*:$;:36<"9$<16*12)":=> Applied Regression Analysis Carlos M. Carvalho How about the original analysis (survey variables)...!""#$%&'$()*+"$,- 58

59 F-test The p-value for the F -test is p-value = Pr(F p,n p 1 > f ) We usually reject the null when the p-value is less than 5%. Big f REJECT! Small p-value REJECT! 59

60 The F-test 4, # 5 ((7 6 # 5 Two equivalent expressions for f 9"3:;$<="<#$3=*;$3";3$4/>$3="$?@A>BA@"C$DA>*AB)";E In Excel, the p-value is reported under Significance F F-tests./0$12/34$45"$/6*7*81)$181)9:*:$;:36<"9$<16*12)":=> Applied Regression Analysis Carlos M. Carvalho!""#$%&'$()*+"$,- 60

61 F-tests The F-test Summary continued ***'./0123"$4$, 8 # ((8 # Note that f is also 4 equal to (you can check the math!), # 5 ((7 6 # 5 SSR/p f = SSE/(n p 1) Two equivalent expressions for f In 9"3:;$<="<#$3=*;$3";3$4/>$3="$?@A>BA@"C$DA>*AB)";E Excel, the values under MS are SSR/p and SSE/(n p 1) Applied Regression Analysis Carlos M. Carvalho f = = 1.39!""#$%&'$()*+"$,- 61

62 Partial F-tests What about fitting a reduced model with only a couple of X s? In other words, do we need all of the X s to explain Y? For example, in the Supervisor data we could argue that X 1 and X 3 were the most important variables in predicting Y. The full model (6 covariates) has Rfull 2 = while the model with only X 1 and X 3 has Rrest 2 = (check that!) Can we make a decision based only in the R 2 calculations? NO!! 62

63 Partial F -test With the total F -test, we were asking Is this regression worthwhile? Now, we re asking Is is useful to add these extra covariates to the regression? You always want to use the simplest model possible. Only add covariates if they are truly informative. 63

64 Partial F -test Consider the regression model Y = β 0 + β 1 X β pbase X pbase + β pbase +1X pbase β pfull X pfull + ε Such that d base is the number of covariates in the base (small) model and p full > p base is the number in the full (larger) model. The Partial F -test is concerned with the hypotheses H 0 : β pbase +1 = β pbase +2 =... = β pfull = 0 H 1 : at least one β j 0 for j > p base. 64

65 Partial F -test It turns out that under the null H 0 (i.e. base model is true), f = (R2 full R2 base )/(p full p base ) (1 Rfull 2 )/(n p full 1) F pfull p base,n p full 1 That is, under the null hypothesis, the ratio of normalized R 2 full R2 base (increase in R 2 ) and 1 R 2 full has F -distribution with p full p base and n p full 1 df. Big f means that R 2 full R2 base is statistically significant. Big f means that at least one of the added X s is useful. 65

66 Supervisor Performance: Partial F -test Back to our supervisor data; we want to test H 0 : β 2 = β 4 = β 5 = β 6 = 0 H 1 : at least one β j 0 for j {2, 4, 5, 6}. The F -stat is f = ( )/(6 2) (1.733)/(30 6 1) = = 0.54 This leads to a p-value of What do we conclude? 66

67 Example: Detecting Sex Discrimination Imagine you are a trial lawyer and you want to file a suit against a company for salary discrimination... you gather the following data... Gender Salary 1 Male Female Female Female Male Female

68 Detecting Sex Discrimination You want to relate salary(y ) to gender(x )... how can we do that? Gender is an example of a categorical variable. The variable gender separates our data into 2 groups or categories. The question we want to answer is: how is your salary related to which group you belong to... Could we think about additional examples of categories potentially associated with salary? MBA education vs. not legal vs. illegal immigrant quarterback vs wide receiver 68

69 Detecting Sex Discrimination We can use regression to answer these question but we need to recode the categorical variable into a dummy variable Gender Salary Sex 1 Male Female Female Female Male Female Note: In Excel you can create the dummy variable using the formula: =IF(Gender= Male,1,0) 69

70 Detecting Sex Discrimination Now you can present the following model in court: How do you interpret β 1? Salary i = β 0 + β 1 Sex i + ɛ i E[Salary Sex = 0] = β 0 β 1 is the male/female difference E[Salary Sex = 1] = β 0 + β 1 70

71 Detecting Sex Discrimination SUMMARY OUTPUT Salary i = β 0 + β 1 Sex i + ɛ i Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 208 ANOVA df SS MS F Significance F Regression E-07 Residual Total Coefficientstandard Erro t Stat P-value Lower 95% Upper 95% Intercept E Gender E ˆβ 1 = b 1 = on average, a male makes approximately $8,300 more than a female in this firm. How should the plaintiff s lawyer use the confidence interval in his presentation? 71

72 Detecting Sex Discrimination How can the defense attorney try to counteract the plaintiff s argument? Perhaps, the observed difference in salaries is related to other variables in the background and NOT to policy discrimination... Obviously, there are many other factors which we can legitimately use in determining salaries: education job productivity experience How can we use regression to incorporate additional information? 72

73 Detecting Sex Discrimination Let s add a measure of experience... Salary i = β 0 + β 1 Sex i + β 2 Exp i + ɛ i What does that mean? E[Salary Sex = 0, Exp] = β 0 + β 2 Exp E[Salary Sex = 1, Exp] = (β 0 + β 1 ) + β 2 Exp 73

74 Detecting Sex Discrimination The data gives us the year hired as a measure of experience... YrHired Gender Salary Sex 1 92 Male Female Female Female Male Female

75 Detecting Sex Discrimination SUMMARY OUTPUT Regression Statistics Multiple R R Square Adjusted R Standard E Observation 208 Salary i = β 0 + β 1 Sex i + β 2 Exp + ɛ i ANOVA df SS MS F Significance F Regression E-31 Residual Total Coefficients tandard Erro t Stat P-value Lower 95% Upper 95% Intercept E Gender E YrHired E Salary i = Sex i 0.98Exp i + ɛ i Is this good or bad news for the defense? 75

76 Detecting Sex Discrimination Salary i = { Expi + ɛ i females Exp i + ɛ i males Salary Year Hired 76

77 More than Two Categories We can use dummy variables in situations in which there are more than two categories. Dummy variables are needed for each category except one, designated as the base category. Why? Remember that the numerical value of each category has no quantitative meaning! 77

78 Example: House Prices We want to evaluate the difference in house prices in a couple of different neighborhoods. Nbhd SqFt Price

79 Example: House Prices Let s create the dummy variables dn1, dn2 and dn3... Nbhd SqFt Price dn1 dn2 dn

80 Example: House Prices Price i = β 0 + β 1 dn1 i + β 2 dn2 i + β 3 Size i + ɛ i E[Price dn1 = 1, Size] = β 0 + β 1 + β 3 Size (Nbhd 1) E[Price dn2 = 1, Size] = β 0 + β 2 + β 3 Size (Nbhd 2) E[Price dn1 = 0, dn2 = 0, Size] = β 0 + β 3 Size (Nbhd 3) 80

81 Example: House Prices Price i = β 0 + β 1 dn1 + β 2 dn2 + β 3 Size + ɛ i SUMMARY OUTPUT Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 128 ANOVA df SS MS F gnificance F Regression E-31 Residual Total Coefficients Standard Error t Stat P-value Lower 95%Upper 95% Intercept dn dn size Price i = dn dn Size + ɛ i 81

82 Example: House Prices Price Nbhd = 1 Nbhd = 2 Nbhd = Size 82

83 Price i = Size + ɛ i 83 Example: House Prices SUMMARY OUTPUT Price i = β 0 + β 1 Size + ɛ i Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 128 ANOVA df SS MS F gnificance F Regression E-11 Residual Total Coefficientsandard Err t Stat P-value ower 95%pper 95% Intercept size

84 Example: House Prices Price Nbhd = 1 Nbhd = 2 Nbhd = 3 Just Size Size 84

85 Back to the Sex Discrimination Case Salary Year Hired Does it look like the effect of experience on salary is the same for males and females? 85

86 Back to the Sex Discrimination Case Could we try to expand our analysis by allowing a different slope for each group? Yes... Consider the following model: Salary i = β 0 + β 1 Exp i + β 2 Sex i + β 3 Exp i Sex i + ɛ i For Females: For Males: Salary i = β 0 + β 1 Exp i + ɛ i Salary i = (β 0 + β 2 ) + (β 1 + β 3 )Exp i + ɛ i 86

87 Sex Discrimination Case How does the data look like? YrHired Gender Salary Sex SexExp 1 92 Male Female Female Female Male Female

88 Sex Discrimination Case Salary i = β 0 + β 1 Sex i + β 2 Exp + β 3 Exp Sex + ɛ i SUMMARY OUTPUT Regression Statistics Multiple R R Square Adjusted R S Standard Err Observations 208 ANOVA df SS MS F Significance F Regression E-45 Residual Total Coefficients tandard Erro t Stat P-value Lower 95% Upper 95% Intercept E Gender E YrHired GenderExp E Salary i = Sex i Exp Exp Sex + ɛ i 88

89 Sex Discrimination Case Salary Year Hired Is this good or bad news for the plaintiff? 89

90 Variable Interaction So, the effect of experience on salary is different for males and females... in general, when the effect of the variable X 1 onto Y depends on another variable X 2 we say that X 1 and X 2 interact with each other. We can extend this notion by the inclusion of multiplicative effects through interaction terms. Y i = β 0 + β 1 X 1i + β 2 X 2i + β 3 (X 1i X 2i ) + ε E[Y X 1, X 2 ] X 1 = β 1 + β 3 X 2 We will pick this up in our next section... 90