addressing alternative explanations: multiple...
TRANSCRIPT
![Page 1: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/1.jpg)
Addressing Alternative Explanations:Explanations: Multiple Regressiong
17.871Spring 2012
![Page 2: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/2.jpg)
Did Cli t h t G lDid Clinton hurt Gore example
Did Clinton hurt Gore in the 2000 election?Treatment is not liking Bill ClintonTreatment is not liking Bill Clinton
How would you test this?
![Page 3: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/3.jpg)
Bivariate regression of Gore thermometer onBivariate regression of Gore thermometer on Clinton thermometer
Clinton thermometer
![Page 4: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/4.jpg)
Did Cli t h t G lDid Clinton hurt Gore example
What alternative explanations would you need to address?
Nonrandom selection into the treatment group (disliking Nonrandom selection into the treatment group (disliking Clinton) from many sources
Let’s address one source: party identification How could we do this?
Matching: compare Democrats who like or don’t like Clinton; do the same for Republicans and independents
Multivariate regression: control for partisanship statistically Also called multiple regression, Ordinary Least Squares (OLS) Presentation below is intuitive
![Page 5: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/5.jpg)
D ti i tDemocratic picture
Clinton thermometer
![Page 6: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/6.jpg)
I d d t i tIndependent picture
Clinton thermometer
![Page 7: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/7.jpg)
R bli i tRepublican picture
Clinton thermometer
![Page 8: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/8.jpg)
C bi d d t i tCombined data picture
Clinton thermometer
![Page 9: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/9.jpg)
Combined data picture withCombined data picture with regression: bias!
Clinton thermometer
![Page 10: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/10.jpg)
Combined data picture withCombined data picture with “true” regression lines overlaid
Clinton thermometer
![Page 11: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/11.jpg)
Tempting yet wrong p g y gnormalizations
Subtract the Goretherm. from theavg. Gore therm.
Clinton thermometer
avg. Gore therm. score
Subtract the ClintonSubtract the Clintontherm. from theavg. Clinton therm.
Clinton thermometer
score
![Page 12: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/12.jpg)
3D R l ti hi3D Relationship
![Page 13: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/13.jpg)
3D Linear Relationship3D Linear Relationship
![Page 14: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/14.jpg)
3D R l ti hi Cli t3D Relationship: Clinton
050100
![Page 15: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/15.jpg)
3D R l ti hi t3D Relationship: party
Rep Ind Dem
![Page 16: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/16.jpg)
The Linear Relationship between ThreeThe Linear Relationship between Three Variables
ClintonGore
XXY
Clinton thermometer
Gorethermometer Party ID
iiii XXY ,22,110
STATA: 1 2reg y x1 x2
reg gore clinton party3
![Page 17: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/17.jpg)
Multivariate slope coefficientsMultivariate slope coefficientsClinton effect (on Gore) in bivariate (B)
vs),cov(ˆ 1 YXB
bivariate (B) regression Are Gore and Party ID
related?
Bivariate estimate:
),cov(ˆ-),cov(ˆ
vs.)var(
)(
212
11
1
11
XXYXX
MM
Bivariate estimate:
Multivariate estimate:)var()var( 1
21
1 XX
Clinton effect ( G ) i
Multivariate estimate:
Are Clinton and Party ID
)cov(ˆ XXMMB ˆˆ
(on Gore) in multivariate (M)
regression
yrelated?
When does ? Obviously, when 0)var(
),cov(ˆ1
212
XXXMMB
11 X1 is Clinton thermometer, X2 is PID, and Y is Gore thermometer
![Page 18: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/18.jpg)
Th Sl C ffi i tThe Slope Coefficients
n
n
iii
n
n
iii XXXXXXYY
1,22,11
21
,11
1 and))((
ˆ-))((
ˆ
nn
n
ii
n
ii XXXX
1
2,11
2
1
2,11
1
)()(
n
n
iii
n
n
iii
XX
XXXX
XX
XXYY
2
1,22,11
12
1,12
2
)(
))((ˆ-
)(
))((ˆ
i
ii
i XXXX1
2,22
1
2,22 )()(
X1 is Clinton thermometer, X2 is PID, and Y is Gore thermometer
![Page 19: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/19.jpg)
The Slope Coefficients MoreThe Slope Coefficients More Simply
and),cov(ˆ),cov(ˆ 211 XXYX and)var(
-)var( 1
21
1 XX
)var(),cov(ˆ-
)var(),cov(ˆ 21
12
2 XXX
XYX
)var()var( 22 XX
X1 is Clinton thermometer, X2 is PID, and Y is Gore thermometer
![Page 20: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/20.jpg)
Th M t i fThe Matrix formy1
y2
1 x1,1 x2,1 … xk,1
1 x1,2 x2,2 … xk,2
…
y
1 … … … …
1 x x xyn 1 x1,n x2,n … xk,n
( )X X X y1 ( ) y
![Page 21: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/21.jpg)
Th O t tThe Output. reg gore clinton party3
Source | SS df MS Number of obs = 1745-------------+------------------------------ F( 2, 1742) = 1048.04
Model | 629261.91 2 314630.955 Prob > F = 0.0000Model | 629261.91 2 314630.955 Prob > F 0.0000Residual | 522964.934 1742 300.209492 R-squared = 0.5461
-------------+------------------------------ Adj R-squared = 0.5456Total | 1152226.84 1744 660.68053 Root MSE = 17.327
------------------------------------------------------------------------------gore | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------clinton | .5122875 .0175952 29.12 0.000 .4777776 .5467975party3 | 5.770523 .5594846 10.31 0.000 4.673191 6.867856p y |_cons | 28.6299 1.025472 27.92 0.000 26.61862 30.64119
------------------------------------------------------------------------------
Interpretation of clinton effect: Holding constant party identification a one-Interpretation of clinton effect: Holding constant party identification, a one-point increase in the Clinton feeling thermometer is associated with a .51 increase in the Gore thermometer.
![Page 22: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/22.jpg)
S t iSeparate regressions
(1) (2) (3)Intercept 23.1 55.9 28.6pClinton 0.62 -- 0.51Party -- 15.7 5.8Party 15.7 5.8
![Page 23: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/23.jpg)
Is the Clinton effect causal?Is the Clinton effect causal? That is, should we be convinced that negative
feelings about Clinton really hurt Gore?feelings about Clinton really hurt Gore? No!
The regression analysis has only ruled out linear d l ti t IDnonrandom selection on party ID.
Nonrandom selection into the treatment could occur from
V i bl th th t ID Variables other than party ID, or Reverse causation, that is, feelings about Gore influencing
feelings about Clinton. Additionally the regression analysis may not have Additionally, the regression analysis may not have
entirely ruled out nonrandom selection even on party ID because it may have assumed the wrong functional form. E.g., what if nonrandom selection on strong
Republican/strong Democrat, but not on weak partisans
![Page 24: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/24.jpg)
Other approaches to addressingOther approaches to addressing confounding effects? Experiments Difference-in-differences designs Difference in differences designs Others?
![Page 25: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/25.jpg)
Summary: Why we controlSummary: Why we control Address alternative explanations by removing
f di ff tconfounding effects Improve efficiency
![Page 26: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/26.jpg)
Why did the Clinton CoefficientWhy did the Clinton Coefficient change from 0.62 to 0.51. corr gore clinton party, cov(obs=1745)
| gore clinton party3-------------+---------------------------+
gore | 660.681clinton | 549.993 883.182
t 3 | 13 7008 16 905 8735party3 | 13.7008 16.905 .8735
![Page 27: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/27.jpg)
Th C l l tiThe Calculations993549)cov(ˆ clintongoreB 6227.0182.883993.549
)var(),cov(
1̂ clinton
clintongoreB
)(),cov(ˆ
)(),cov(ˆ
21 li t
partyclintonli t
clintongore MM
905.167705.5993.549)var()var( 21
clintonclinton
1105.06227.0182.883182.883
. corr gore clinton party,cov(obs=1745)
| gore clinton party3
5122.0 -------------+---------------------------gore | 660.681
clinton | 549.993 883.182party3 | 13.7008 16.905 .8735
![Page 28: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/28.jpg)
D i ki d G k Lif E lDrinking and Greek Life Example
Why is there a correlation between living in a fraternity/sorority house and drinking?y y gGreek organizations often emphasize social
gatherings that have alcohol. The effect is g gbeing in the Greek organization itself, not the house.
There’s something about the House environment itself.
![Page 29: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/29.jpg)
Dependent variable: TimesDependent variable: Times Drinking in Past 30 Days
![Page 30: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/30.jpg)
. infix age 10-11 residence 16 greek 24 screen 102 timespast30 103 howmuchpast30 104 gpa 278-279 studying 281 i h 32 h hh 326 i li i 283 99 93timeshs 325 howmuchhs 326 socializing 283 stwgt 99 475-493weight99 494-512 using da3818.dat,clear(14138 observations read)
. recode timespast30 timeshs (1=0) (2=1.5) (3=4) (4=7.5) (5=14.5) (6=29.5) (7=45)(timespast30: 6571 changes made)(timeshs 10272 changes made)(timeshs: 10272 changes made)
. replace timespast30=0 if screen<=3(4631 real changes made)
![Page 31: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/31.jpg)
. tab timespast30
timespast30 | Freq. Percent Cum.+------------+-----------------------------------
0 | 4,652 33.37 33.371.5 | 2,737 19.64 53.014 | 2,653 19.03 72.04
7.5 | 1,854 13.30 85.3414.5 | 1,648 11.82 97.1729.5 | 350 2.51 99.6845 | 45 0 32 100 0045 | 45 0.32 100.00
------------+-----------------------------------Total | 13,939 100.00
![Page 32: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/32.jpg)
K l t i blKey explanatory variables
Live in fraternity/sorority house Indicator variable (dummy variable) Indicator variable (dummy variable) Coded 1 if live in, 0 otherwise
Member of fraternity/sorority Member of fraternity/sorority Indicator variable (dummy variable)Coded 1 if member 0 otherwiseCoded 1 if member, 0 otherwise
![Page 33: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/33.jpg)
Th R iThree RegressionsDependent variable: number of times drinking in past 30 daysDependent variable: number of times drinking in past 30 days
Live in frat/sor house (indicator variable)
4.44(0.35)
--- 2.26(0.38)( ) ( )
Member of frat/sor (indicator variable)
--- 2.88(0.16)
2.44(0.18)
Intercept 4 54 4 27 4 27Intercept 4.54(0.56)
4.27(0.059)
4.27(0.059)
R2 .011 .023 .025
N 13,876 13,876 13,876
N t St d d i th C B t li i iWhat is the substantive interpretation of the coefficients?Note: Standard errors in parentheses. Corr. Between living in frat/sor house and being a member of a Greek organization is .42
![Page 34: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/34.jpg)
Th Pi tThe Picture
i kLiving in frat house
=2.26
X
M2̂
Drinks per 30 days
house
=0.1921Y
X2
Member of fraternity =2.44
X1
M1
ˆ y 1
![Page 35: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/35.jpg)
A ti f th t t l ff tAccounting for the total effect
ˆˆˆ21211
ˆˆ ˆ MMB Total effect = Direct effect + indirect effect
Drinks perLiving in frat house
=2.26
X2
M2̂
Drinks per 30 days
=0.1921Y
Member of fraternity =2.44
X1
M1̂
![Page 36: Addressing Alternative Explanations: Multiple Regressionweb.mit.edu/~17.871/www/2012/04multiple_regression_2012.pdf · Combined data picture withCombined data picture with regression:](https://reader034.vdocuments.net/reader034/viewer/2022052007/601c06270fac17342a41b093/html5/thumbnails/36.jpg)
Accounting for the effects of frat ghouse living and Greek membership on drinkingmembership on drinking
Effect Total Direct IndirectMember of 2.88 2.44 0.44Greek org. (85%) (15%)Live in frat/ 4.44 2.26 2.18sor. house (51%) (49%)