Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 5

ANSWER KEY

Chapter 8, Exercise 6

(a) Estimate a pooled OLS model with deaths per mile as the dependent variable and cell
phone ban and text ban as the two independent variables. Briefly interpret the results.
Answer:
Both variables are highly statistically significantly associated with decrease
in deaths per mile.
reg DeathsPerBillionMiles cell_ban text_ban
Source | SS df MS Number of obs = 306
-------------+------------------------------ F( 2, 303) = 49.62
Model | 880.932526 2 440.466263 Prob > F = 0.0000
Residual | 2689.50557 303 8.87625599 R-squared = 0.2467
-------------+------------------------------ Adj R-squared = 0.2418
Total | 3570.43809 305 11.7063544 Root MSE = 2.9793
------------------------------------------------------------------------------
DeathsPerB~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
cell_ban | -2.838609 .5364153 -5.29 0.000 -3.89418 -1.783039
text_ban | -2.061566 .4066788 -5.07 0.000 -2.861838 -1.261294
_cons | 13.38078 .2094036 63.90 0.000 12.96871 13.79285
------------------------------------------------------------------------------

(b) Describe a possible state fixed effect that could cause endogeneity and bias in the model
from part (a).
Answer:
Failure to control for state fixed effect can cause bias if there are state fixed
effects (meaning numbers of deaths vary from state to state across all years) and
that if these fixed effects are correlated with cell phone and texting bans. For the
fixed effects to exist means that the some states have safer roads (fewer deaths) or
more dangerous roads. It is quite plausible that the states with regulations on cell
phones and texting, also have more regulations that lead to road safety (think
Minnesota) or spend more on roads or have inherently more cautious citizens (think
Utah).

(c) Estimate a one-way fixed effects model that controls for state-level fixed effects. Include
deaths per mile as the dependent variable and cell phone ban and text ban as the two
independent variables. Does the coefficient on cell phone ban change in the manner you
would expect based on your answer from part (a)?
Answer:
The coefficient on cell phone ban drops from -2.8 to -0.81 ; the coefficient on
text bans drops from -2.06 to -1.12. The cell phone ban coefficient is not statistically
significant at the = 0.10 (two-sided) level, but the text ban variable is. The drop in
coefficient size for both variables makes sense as it is likely that the kind of states
that had cell phone and text bans also had other policies or behaviors associated
with lower death rates.
xtreg DeathsPerBillionMiles cell_ban text_ban, fe i(state_numeric)
Fixed-effects (within) regression Number of obs = 306
Group variable: state_nume~c Number of groups = 51

R-sq: within = 0.1370 Obs per group: min = 6


between = 0.3337 avg = 6.0
overall = 0.2382 max = 6
F(2,253) = 20.08
corr(u_i, Xb) = 0.3305 Prob > F = 0.0000
------------------------------------------------------------------------------
DeathsPerB~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
cell_ban | -.8186268 .5195726 -1.58 0.116 -1.841865 .2046116
text_ban | -1.125641 .2193892 -5.13 0.000 -1.557703 -.6935796
_cons | 12.76515 .1144803 111.51 0.000 12.53969 12.99061
-------------+----------------------------------------------------------------
sigma_u | 2.9126192
sigma_e | 1.3381466
rho | .82571163 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(50, 253) = 27.53 Prob > F = 0.0000

(d) Describe a possible year fixed effect that could cause endogeneity and bias in the fixed
effects model in part (c).
Answer:
Failure to control for year fixed effect can cause bias if there are year fixed
effects (meaning numbers of deaths vary from year to year across all states) and
that if these fixed effects are correlated with cell phone bans. For the fixed effects to
exist means that the roads are getting safer (fewer deaths in recent years) or more
dangerous (more deaths in recent years) or even that there is simply variation
across years (some years have many deaths and others have fewer, across all states).
If cell phone bans vary across years systematically (as they very likely do) then it is
possible for cell phone ban variables to correlate with omitted year fixed effects
which can cause bias.
A plausible story (consistent with what we find below) is that fatalities are
decreasing over time on average across all states (meaning there are year fixed
effects), and that cell bans are increasing over time (as there are more cell phone
bans in recent years).

(e) Use the hybrid de-meaned approach discussed in the chapter to estimate a two-way fixed
effects model. Include deaths per mile as the dependent variable and cell phone ban and
text ban as the two independent variables. Does the coefficient on cell phone ban change
in the manner you would expect based on your answer in part (d)?
Answer:
The coefficient on cell phone bans falls to -0.61. The coefficient flips to being
positive. Both changes are consistent with the possibility that cell phone and texting
bans were more common in more recent years and that more recent years have been
safer.
xtreg DeathsPerBillionMiles cell_ban text_ban i.year, fe i(state_numeric)
Fixed-effects (within) regression Number of obs = 306
Group variable: state_nume~c Number of groups = 51
R-sq: within = 0.4962 Obs per group: min = 6
between = 0.0807 avg = 6.0
overall = 0.0962 max = 6
F(7,248) = 34.89
corr(u_i, Xb) = 0.0398 Prob > F = 0.0000
------------------------------------------------------------------------------
DeathsPerB~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
cell_ban | -.6008733 .4019314 -1.49 0.136 -1.392508 .1907609
text_ban | .2913469 .2201367 1.32 0.187 -.142229 .7249227
year |
2008 | -1.147373 .2046891 -5.61 0.000 -1.550523 -.7442219
2009 | -1.937611 .2067455 -9.37 0.000 -2.344812 -1.530411
2010 | -2.465837 .2193803 -11.24 0.000 -2.897923 -2.033751
2011 | -2.749958 .2378047 -11.56 0.000 -3.218333 -2.281584
2012 | -2.468676 .2468671 -10.00 0.000 -2.954899 -1.982452
_cons | 14.055 .1495555 93.98 0.000 13.76043 14.34956
-------------+----------------------------------------------------------------
sigma_u | 3.144867
sigma_e | 1.0327095
rho | .9026631 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(50, 248) = 43.89 Prob > F = 0.0000

We can confirm our intuition by looking at deaths by year and the cell phone
and text bans over the years. The death rate was falling over the time period of the
data while the proportion of states with the two kinds of bans was rising over the
time period .
(In Stata, you can see this with the following commands:
by year, sort : summarize DeathsPerBillionMiles
by year, sort : summarize cell_ban
by year, sort : summarize text_ban )
As a general matter, we need not run such additional analyses. We simply
can note the possibility of bias and run a fixed effects model. In this problem we
show these additional steps to help understand what is going on.
Note that the direction of bias from not including year fixed effects is not
something that we can immediately or even certainly know. Instead, the point is to
think about what could be happening and how our model could off-set that. In this
case, we can, if we want, do a bit more investigating to see that, indeed, cell phone
bans became more common and that deaths declined. While there is no reason we
would know this right away without looking at the data or having outside
knowledge about traffic deaths and cell bans, we could say that if there is a pattern
of increasing or decreasing cell phone bans over time and increasing or decreasing
deaths over time, we could have bias if we omit year fixed effects.

(f) The model in part (e) is somewhat sparse with regard to control variables. Estimate a
two-way fixed effects model that includes control variables for cell phones per 10,000
people and percent urban. Briefly describe changes in inference about the effect of cell
phone and text bans.
Answer:
The results for the total miles driven and cell phone bans are similar as
above, with the cell ban being associated decrease of 0.67 deaths per billion miles
drive. This is statistically significant at =0.10 (two-sided). The coefficient on text
ban is 0.25, but is not statistically significant. Cell phones subscriptions (per 10,000)
are associated with fewer deaths.
xtreg DeathsPerBillionMiles cell_ban text_ban cell_per10thous_pop urban_percent i.year ,
fe i(state_numeric)
Fixed-effects (within) regression Number of obs = 306
Group variable: state_nume~c Number of groups = 51

R-sq: within = 0.5065 Obs per group: min = 6


between = 0.0065 avg = 6.0
overall = 0.0750 max = 6
F(9,246) = 28.05
corr(u_i, Xb) = -0.0438 Prob > F = 0.0000
-------------------------------------------------------------------------------------
DeathsPerBillionM~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------------+----------------------------------------------------------------
cell_ban | -.6797957 .4029491 -1.69 0.093 -1.473466 .1138748
text_ban | .2559262 .2221923 1.15 0.251 -.1817158 .6935682
cell_per10thous_pop | -.0003404 .0001729 -1.97 0.050 -.000681 2.59e-07
urban_percent | .0131348 .0111986 1.17 0.242 -.0089226 .0351922
year |
2008 | -1.006156 .2158168 -4.66 0.000 -1.431241 -.5810718
2009 | -1.663634 .2479518 -6.71 0.000 -2.152014 -1.175255
2010 | -2.116744 .2868242 -7.38 0.000 -2.681689 -1.5518
2011 | -2.268126 .3423523 -6.63 0.000 -2.942442 -1.593811
2012 | -1.927766 .3712164 -5.19 0.000 -2.658934 -1.196598
_cons | 16.36945 1.448329 11.30 0.000 13.51674 19.22215
--------------------+----------------------------------------------------------------
sigma_u | 3.1882565
sigma_e | 1.0262163
rho | .90612315 (fraction of variance due to u_i)
-------------------------------------------------------------------------------------
F test that all u_i=0: F(50, 246) = 36.63 Prob > F = 0.0000

(g) Estimate the same two-way fixed effects model by using the least squares dummy
variable (LSDV) approach. Compare the coefficient and t statistic on the cell phone
variable to the results from part (f).
Answer:
The coefficient, standard error, t statistic and confidence intervals on the
coefficient on cell_ban is identical to the estimate from the hybrid two-way fixed
effect model above.
reg DeathsPerBillionMiles cell_ban text_ban cell_per10thous_pop urban_percent i.year
i.state_numeric
Source | SS df MS Number of obs = 306
-------------+------------------------------ F( 59, 246) = 53.29
Model | 3311.37061 59 56.1249255 Prob > F = 0.0000
Residual | 259.067485 246 1.05311986 R-squared = 0.9274
-------------+------------------------------ Adj R-squared = 0.9100
Total | 3570.43809 305 11.7063544 Root MSE = 1.0262
-------------------------------------------------------------------------------------
DeathsPerBillionM~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------------+----------------------------------------------------------------
cell_ban | -.6797957 .4029491 -1.69 0.093 -1.473466 .1138748
text_ban | .2559262 .2221923 1.15 0.251 -.1817158 .6935682
cell_per10thous_pop | -.0003404 .0001729 -1.97 0.050 -.000681 2.59e-07
urban_percent | .0131348 .0111986 1.17 0.242 -.0089226 .0351922
year |
2008 | -1.006156 .2158168 -4.66 0.000 -1.431241 -.5810718
2009 | -1.663634 .2479518 -6.71 0.000 -2.152014 -1.175255
2010 | -2.116744 .2868242 -7.38 0.000 -2.681689 -1.5518
2011 | -2.268126 .3423523 -6.63 0.000 -2.942442 -1.593811
2012 | -1.927766 .3712164 -5.19 0.000 -2.658934 -1.196598
state_numeric |
2 | -1.736801 .639553 -2.72 0.007 -2.9965 -.4771029
3 | -.9479356 .6202631 -1.53 0.128 -2.169639 .2737682
[ ** multiple dummy variables omitted to save space ** ]
50 | -4.652387 .6175067 -7.53 0.000 -5.868662 -3.436113
51 | .6863052 .6397375 1.07 0.284 -.5737564 1.946367
|
_cons | 18.87526 1.494318 12.63 0.000 15.93197 21.81855
-------------------------------------------------------------------------------------

(h) Based on the LSDV results, identify states with large positive and negative fixed effects.
Explain what these mean (being sure to note the excluded category) and speculate about
how the positive and negative fixed effect states differ. (It is helpful to connect the state
number to state name; in Stata, do this with the command list state state_numeric if year
==2012.)
Answer:
The excluded category is state_numeric =1 which is Alabama. Hence the
fixed effects indicate how different each state is from Alabama. States that have
very small, insignificant fixed effects have fixed effect similar to Alabama’s (these
include state 37 (Oklahoma), state 42 (South Dakota), state 43 (Tennessee ) and state
51 (Wyoming). States with positive and significant fixed effects have more deaths
per billion miles, controlling for the variables in the model. In other words, these
states were more dangerous for the whole panel. They include state 4 (Arkansas),
state 18 (Kentucky), 19(Louisiana) , state 27(Montana) and state 41 (South
Carolina) .
States that have large negative fixed effects were less dangerous for the whole
panel. Presumably they had more investment, regulation and other behaviors that
would be associated with safer roads. There are a large number of such states, but
some of the bigger negative fixed effects were for state 7 (Connecticut), state 22
(Massachusetts), state 24 (Minnesota) , state 31 (New Jersey) and state 40 (Rhode
Island).

You might also like