ECON4150 - Introductory Econometrics

Seminar 6

Monique de Haan
([email protected])

Stock and Watson Chapter 10

 2

Empirical exercise E10.1: Data

• Some U.S. states have enacted ’shall-issue’ laws which allow citizens to
carry concealed weapons.

• We are going to investigate the effect of shall-issue laws on violent

crime rates.

• In this exercise we use the data set Guns.dta.

• This is a balanced panel of data on 50 US states, plus the District of

Columbia for the years 1977 – 1999.

• There are a total of 51 states × 23 years = 1173 observations.

• These data were provided by Professor John Donohue of Stanford

University.

• Data were used in the: “Shooting Down the ‘More Guns Less Crime’
Hypothesis” Stanford Law Review (2003)
 3
Review, 2003, Vol. 55, 1193-1312.
Empirical exercise E10.1: Data

Variable Definitions

Variable Definition
vio violent crime rate (incidents per 100,000 members of the population)
rob robbery rate (incidents per 100,000)
mur murder rate (incidents per 100,000)
shall = 1 if the state has a shall-carry law in effect in that year
= 0 otherwise
incarc_rate incarceration rate in the state in the previous year (sentenced
prisoners per 100,000 residents; value for the previous year)
density population per square mile of land area, divided by 1000
avginc real per capita personal income in the state, in thousands of dollars
pop state population, in millions of people
pm1029 percent of state population that is male, ages 10 to 29
pw1064 percent of state population that is white, ages 10 to 64
pb1064 percent of state population that is black, ages 10 to 64
stateid ID number of states (Alabama = 1, Alaska = 2, etc.)
year Year (1977-1999)
 4

Empirical exercise E10.1: Data

Friday February 28 14:36:14 2014 Page 1

___ ____ ____ ____ ____(R)

/__ / ____/ / ____/
___/ / /___/ / /___/
Statistics/Data Analysis

1 . sum vio mur rob shall incarc_rate pb1064 pw1064 pm1029 pop avginc density stateid year

Variable Obs Mean Std. Dev. Min Max

vio 1173 503.0747 334.2772 47 2921.8

mur 1173 7.665132 7.52271 .2 80.6
rob 1173 161.8202 170.51 6.4 1635.1
shall 1173 .2429668 .4290581 0 1
incarc_rate 1173 226.5797 178.8881 19 1913

pb1064 1173 5.336217 4.885688 .2482066 26.97957

pw1064 1173 62.94543 9.761527 21.78043 76.52575
pm1029 1173 16.08113 1.732143 12.21368 22.35269
pop 1173 4.816341 5.252115 .402753 33.14512
avginc 1173 13.7248 2.554543 8.554884 23.64671

density 1173 .3520382 1.355472 .0007071 11.10212

stateid 1173 28.96078 15.68352 1 56
year 1173 88 6.636079 77 99
 5

Empirical exercise E10.1: part (a)

Friday February 28 14:42:51 2014 Page 1

___ ____ ____ ____ ____(R)

/__ / ____/ / ____/
___/ / /___/ / /___/
Statistics/Data Analysis

1 . gen ln_vio=ln(vio)

2 . regress ln_vio shall, robust

Linear regression Number of obs = 1173

F( 1, 1171) = 86.86
Prob > F = 0.0000
R-squared = 0.0866
Root MSE = .61735

Robust
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]

shall -.4429646 .0475283 -9.32 0.000 -.5362148 -.3497144

_cons 6.134919 .0193039 317.81 0.000 6.097045 6.172793
 Friday February 28 14:43:54 2014 Page 1 6

Empirical exercise E10.1: part (a) ___ ____ ____ ____ ____(R)
/__ / ____/ / ____/
___/ / /___/ / /___/
Statistics/Data Analysis

1 . regress ln_vio shall incarc_rate density avginc pop pb1064 pw1064 pm1029, robust

Linear regression Number of obs = 1173

F( 8, 1164) = 95.67
Prob > F = 0.0000
R-squared = 0.5643
Root MSE = .42769

Robust
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]

shall -.3683869 .0347879 -10.59 0.000 -.436641 -.3001329

incarc_rate .0016126 .0001807 8.92 0.000 .0012581 .0019672
density .0266885 .0143494 1.86 0.063 -.0014651 .054842
avginc .0012051 .0072778 0.17 0.869 -.013074 .0154842
pop .0427098 .0031466 13.57 0.000 .0365361 .0488836
pb1064 .0808526 .0199924 4.04 0.000 .0416274 .1200778
pw1064 .0312005 .0097271 3.21 0.001 .012116 .0502851
pm1029 .0088709 .0120604 0.74 0.462 -.0147917 .0325334
_cons 2.981738 .6090198 4.90 0.000 1.786839 4.176638

• The coefficient equals -0.368, which suggests that shall-issue laws

reduce the violent crime rate by 36%.
• This is a large effect.
 2 . regress ln_vio shall, robust
7
Linear regression Number of obs = 1173
Empirical exercise E10.1: part (a) F( 1, 1171) = 86.86
Prob > F = 0.0000
Friday February 28 14:43:54 2014 Page 1
R-squared = 0.0866
(1) Regression without control variables: Root MSE =
___ ____ ____ ____ ____(R)
.61735
/__ / ____/ / ____/
___/ / /___/ / /___/
Robust
Statistics/Data Analysis
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]
1 . regress ln_vio shall incarc_rate density avginc pop pb1064 pw1064 pm1029, robust
shall -.4429646 .0475283 -9.32 0.000 -.5362148 -.3497144
_cons 6.134919 .0193039 317.81 0.000 6.097045 6.172793
Linear regression Number of obs = 1173
F( 8, 1164) = 95.67
Prob > F = 0.0000
R-squared = 0.5643
(2) Regression with control variables Root MSE = .42769

Robust
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]

shall -.3683869 .0347879 -10.59 0.000 -.436641 -.3001329

incarc_rate .0016126 .0001807 8.92 0.000 .0012581 .0019672
density .0266885 .0143494 1.86 0.063 -.0014651 .054842
• The coefficient
avginc .0012051 βshall = −0.443;
in (1) is.0072778
b 0.17 in 0.869
(2) it is βshall
b = −0.369.
-.013074 Both
.0154842
pop .0427098 .0031466 13.57 0.000 .0365361 .0488836
are highly statistically
pb1064 .0808526 significant.
.0199924 4.04 0.000 .0416274 .1200778
pw1064 .0312005 .0097271 3.21 0.001 .012116 .0502851
.0088709 .0120604
pm1029
• Adding
_cons the control
2.981738variables
.6090198results 0.74
in a
4.90
0.462
small
0.000drop -.0147917
in .0325334
the estimated
1.786839 4.176638
coefficient.
• Possible omitted variables that vary between states but not over time:
Attitudes towards guns and crime, quality of police and other
crime-prevention programs.
 ___ ____ ____ ____ ____(R) 8
/__ / ____/ / ____/
Empirical exercise E10.1: part (b) ___/ / /___/ / /___/
Statistics/Data Analysis

1 . xtset state
panel variable: stateid (balanced)

2 . xtreg ln_vio shall incarc_rate density avginc pop pb1064 pw1064 pm1029, fe robust

Fixed-effects (within) regression Number of obs = 1173

Group variable: stateid Number of groups = 51

R-sq: within = 0.2178 Obs per group: min = 23

between = 0.0033 avg = 23.0
overall = 0.0001 max = 23

F(8,50) = 34.10
corr(u_i, Xb) = -0.3687 Prob > F = 0.0000

(Std. Err. adjusted for 51 clusters in stateid)

Robust
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]

shall -.0461415 .0417616 -1.10 0.275 -.1300222 .0377392

incarc_rate -.000071 .0002504 -0.28 0.778 -.0005739 .0004318
density -.17229 .1376128 -1.25 0.216 -.4486935 .1041135
avginc -.0092037 .0129649 -0.71 0.481 -.0352445 .016837
pop .0115247 .014224 0.81 0.422 -.0170452 .0400945
pb1064 .1042804 .0326849 3.19 0.002 .0386308 .16993
pw1064 .0408611 .0134585 3.04 0.004 .0138289 .0678932
pm1029 -.0502725 .0206949 -2.43 0.019 -.0918394 -.0087057
_cons 3.866017 .7701057 5.02 0.000 2.319214 5.41282

sigma_u .68024947
sigma_e .16072287
rho .94712778 (fraction of variance due to u_i)
 9

Empirical exercise E10.1: part (b)

• The results change when we add state fixed effects.

• The absolute value of the coefficient on shall falls to 0.046, a large

reduction in the coefficient from 0.369 without fixed effects.

• Evidently there was important omitted variable bias in the specification

without fixed effects.

• The estimate of the effect of shall issue laws on the violent crime rate is
no longer statistically significantly different from zero

• The regression model with fixed effects is more credible because this
controls for unobserved characteristics that vary between states but that
are constant over time
1 . xi: xtreg ln_vio shall i.year incarc_rate density avginc pop pb1064 pw1064 pm1029, fe robust
i.year _Iyear_77-99 (naturally coded; _Iyear_77 omitted)

Fixed-effects (within) regression Number of obs = 1173

Group variable: stateid Number of groups = 51

R-sq: within = 0.4180 Obs per group: min = 23

between = 0.0419 avg = 23.0
overall = 0.0009 max = 23

F(30,50) = 56.86
corr(u_i, Xb) = -0.2929 Prob > F = 0.0000

(Std. Err. adjusted for 51 clusters in stateid)

Robust
ln_vio Coef. Std. Err. t P>|t| [95% Conf. Interval]

shall -.0279935 .0407168 -0.69 0.495 -.1097757 .0537886

_Iyear_78 .0585261 .0161556 3.62 0.001 .0260767 .0909755
_Iyear_79 .1639486 .0244579 6.70 0.000 .1148234 .2130738
_Iyear_80 .2170759 .0334184 6.50 0.000 .1499531 .2841987
_Iyear_81 .2172551 .0391956 5.54 0.000 .1385284 .2959819
_Iyear_82 .1946328 .0465743 4.18 0.000 .1010856 .28818
_Iyear_83 .158645 .0593845 2.67 0.010 .0393676 .2779223
_Iyear_84 .1929883 .0770021 2.51 0.015 .0383251 .3476515
_Iyear_85 .2444765 .0922217 2.65 0.011 .0592438 .4297091
_Iyear_86 .3240904 .1089181 2.98 0.004 .1053219 .5428589
_Iyear_87 .324365 .1249881 2.60 0.012 .0733189 .5754111
_Iyear_88 .3867412 .1397074 2.77 0.008 .1061305 .6673518
_Iyear_89 .4422142 .1535358 2.88 0.006 .1338286 .7505999
_Iyear_90 .5430478 .1960859 2.77 0.008 .1491976 .936898
_Iyear_91 .5959456 .2040685 2.92 0.005 .1860619 1.005829
_Iyear_92 .6275171 .2170306 2.89 0.006 .1915982 1.063436
_Iyear_93 .6497414 .2246177 2.89 0.006 .1985834 1.100899
_Iyear_94 .6354187 .2332437 2.72 0.009 .1669349 1.103903
_Iyear_95 .6276831 .2423607 2.59 0.013 .1408874 1.114479
_Iyear_96 .5713423 .2534067 2.25 0.029 .06236 1.080325
_Iyear_97 .5501153 .2613516 2.10 0.040 .0251751 1.075056
_Iyear_98 .4932905 .2746546 1.80 0.079 -.0583696 1.044951
_Iyear_99 .4328777 .2862198 1.51 0.137 -.1420116 1.007767
incarc_rate .000076 .0002079 0.37 0.716 -.0003416 .0004935
density -.0915549 .1238622 -0.74 0.463 -.3403395 .1572297
 11

Empirical exercise E10.1: part (c)

• The absolute value of the coefficient on shall falls further to 0.028, the
Sunday March 2 15:23:09 2014 Page 1
coefficient is not significantly different from zero.
___ ____ ____ ____ ____(R)
• The time effects are jointly statistically significant,
/__
___/ /
/ ____/
so this//regression
/___/
____/
/___/
Statistics/Data Analysis
seems better specified than the regression in part (b).
1 . test _Iyear_78= _Iyear_79= _Iyear_80= _Iyear_81= _Iyear_82= _Iyea
> r_83= _Iyear_84= _Iyear_85= _Iyear_86= _Iyear_87= _Iyear_88= _Iye
> ar_89= _Iyear_90= _Iyear_91= _Iyear_92= _Iyear_93= _Iyear_94= _Iy
> ear_95= _Iyear_96= _Iyear_97= _Iyear_98= _Iyear_99=0

( 1) _Iyear_78 - _Iyear_79 = 0
( 2) _Iyear_78 - _Iyear_80 = 0
( 3) _Iyear_78 - _Iyear_81 = 0
( 4) _Iyear_78 - _Iyear_82 = 0
( 5) _Iyear_78 - _Iyear_83 = 0
( 6) _Iyear_78 - _Iyear_84 = 0
( 7) _Iyear_78 - _Iyear_85 = 0
( 8) _Iyear_78 - _Iyear_86 = 0
( 9) _Iyear_78 - _Iyear_87 = 0
(10) _Iyear_78 - _Iyear_88 = 0
(11) _Iyear_78 - _Iyear_89 = 0
(12) _Iyear_78 - _Iyear_90 = 0
(13) _Iyear_78 - _Iyear_91 = 0
(14) _Iyear_78 - _Iyear_92 = 0
(15) _Iyear_78 - _Iyear_93 = 0
(16) _Iyear_78 - _Iyear_94 = 0
(17) _Iyear_78 - _Iyear_95 = 0
(18) _Iyear_78 - _Iyear_96 = 0
(19) _Iyear_78 - _Iyear_97 = 0
(20) _Iyear_78 - _Iyear_98 = 0
(21) _Iyear_78 - _Iyear_99 = 0
(22) _Iyear_78 = 0

F( 22, 50) = 21.62

Prob > F = 0.0000
 12

Empirical exercise E10.1: part (d)

Dependent variable is ln(rob)

(1) (2) (3) (4)
shall -0.773*** -0.529*** -0.008 0.027
(0.069) (0.051) (0.055) (0.052)
Control variables - yes yes yes
State fixed effects - - yes yes
Time fixed effects - - - yes
Dependent variable is ln(rob)
(1) (2) (3) (4)
shall -0.473*** -0.313*** -0.061 -0.015
(0.049) (0.036) (0.037) (0.038)
Control variables - yes yes yes
State fixed effects - - yes yes
Time fixed effects - - - yes

The results are similar to the results using violent crimes:

• There is a large estimated effect of concealed weapons laws in
specifications (1) and (2).
• This effect is however due to omitted variable bias because the effect
disappears when state and time effects are added.
 13

Empirical exercise E10.1: part (e)

Remaining threats to internal validity:

Omitted variables: There might be important variables that vary between

states and over time that are omitted from the regression
model. For example other policy measures that are related to
the implementation of shall issue laws and that affect crime
rates.

Simultaneous causality: If there are many violent crimes this may induce
policy makers to change concealed weapons laws.
 14

Empirical exercise E10.1: part (f)

• The most credible results include both state fixed effects and time fixed
effects.

• These results indicate that there is no significant effect of concealed

weapon laws on the violent crime rate, the robbery rate nor on the
murder rate.