Вы находитесь на странице: 1из 22

INSTITUTE OF BUSINESS ADMINISTRATION

Speed limit impact on


accidents in U.S
Ahsan Ahmad, Hassan Saeed, Muhammad Khizer Alam, Hasham Zahid, Osama Yousuf, Mahmood
Sarwar khan, Asad Jamil
4/18/2016

[The aim of this paper is to ascertain the effect of speed limit on number of accidents on rural roads of
U.S using two different models of regression and using two different statistical softwares. The two
different methods yield two different results. It is up to a statisian to pick either of the methods. Passing
judgment on any model is not the purpose of this paper. However as authors we prefer the first
method]
Speed limit impact on accidents in U.S 2016

Acknowledgement
First and foremost, we would like to pay our utmost Gratitude to Almighty Allah for blessing us with all
the health, strength and courage to carry out this whole process.
We would also like to thank our course instructor, Sir Muhammad Abdus Salam for his support,
guidance and encouragement throughout the semester to proceed further with the project.

2|Page
Speed limit impact on accidents in U.S 2016

Table of Contents
Introduction .................................................................................................................................................. 4
Resources Used ............................................................................................................................................. 4
Explanation of Variables ............................................................................................................................... 5
Analysis ......................................................................................................................................................... 6
Method 1: ............................................................................................................................................. 6
Method No. 2 ...................................................................................................................................... 11
Unit Root Test For RTOTACC ............................................................................................................... 12
Unit Root Test For UNEM .................................................................................................................... 13
Unit Root Test For RTOTACC with first difference .............................................................................. 15
Unit Root Test For Unem with first difference ................................................................................... 17
Conclusion ................................................................................................................................................... 21
Bibliography ................................................................................................................................................ 22

3|Page
Speed limit impact on accidents in U.S 2016

Introduction
In 1986 the state of California raised its maximum speed limit from 55 miles per hour to 65 miles per
hour. There were numerous studies conducted to find whether the increase in the speed law has
resulted in any increase in the number of accidents in the rural areas. Some studies such as those of
Arthur Van Benthems claimed that the new speed law has been a major factor contributing to the
increase in the number of accidents1; however there were some studies which claimed otherwise. Our
aim is to investigate the general association of increase in speed limit on the number of accidents on
rural road.

To test this theory we have taken the null hypothesis that the new speed law does not have an impact
on number of rural accidents and alternate hypothesis - the speed law has resulted in an increase in the
number of accidents.

HO = Speed law does not have any effect on the number of rural accidents

HA=Speed law has a an impact on the number of rural accidents

We have taken the total accidents in rural areas (rtotacc) as dependent variable and speed law (spdlaw)
as independent variable. Furthermore, we have added some additional variables such as seat belt law
(beltlaw), as well as unemployment (unem) and weekends (wkends). In addition we have included the
months to find out if accidents were high during the vacation season.

Resources Used
The data for the report has been collected from Problem C10.11 from Introductory Econometrics: A
Modern Approach by By Jeffrey M. Wooldridge. We have used a total of 108 observations regarding
automobile accidents, traffic rule, weekends and other variables for the period from January 1981 to
December 1989.

The software which we have used for our report are Strata and Eviews along with MS Excel

1
Do We Need Speed Limits on Freeways? Arthur van Benthem
http://www.economics.uci.edu/files/news_events/2012/VanBentham_jobpaper.pdf

4|Page
Speed limit impact on accidents in U.S 2016

Explanation of Variables

Variables Used Explanation of Variable


Rtotacc total number of rural accidents at 65 mph
Spdlaw implementation of speed law, 1 when speed law has been implemented
Unem Unemployement rate
Stdiffrtot total number of rural accidents at 65 mph (first difference)
stdiffrtotunemp Unemployement rate (first difference)
Beltlaw implementation of belt law, 1 when speed law has been implemented
Wkends number of weekends in a month
T time variable
Feb When the current month is February, 1 when it is February
Mar When the current month is March, 1 when it is March
Apr When the current month is April, 1 when it is April
May When the current month is May, 1 when it is May
Jun When the current month is June, 1 when it is June
Jul When the current month is July, 1 when it is July
Aug When the current month is August, 1 when it is August
Sep When the current month is September, 1 when it is September
Oct When the current month is October, 1 when it is October
nov When the current month is November, 1 when it is November
Dec When the current month is December, 1 when it is December

5|Page
Speed limit impact on accidents in U.S 2016

Analysis
There can be two methods to analyze time series data: 1). Introduce a time independent variable along
with seasonal dummies 2). First differencing method with seasonal dummies

Method 1:
We first ran the simple regression test with time (t) to account for time series on the data which gave
the following result:
Source SS df MS Number of obs = 107
F( 2, 104) = 89.19
Model 438102.209 2 219051.105 Prob > F = 0.0000
Residual 255430.408 104 2456.06161 R-squared = 0.6317
Adj R-squared = 0.6246
Total 693532.617 106 6542.76054 Root MSE = 49.559

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

spdlaw 64.66963 17.07464 3.79 0.000 30.80997 98.52929


t 1.240328 .2507733 4.95 0.000 .7430349 1.737621
_cons 237.5105 11.20407 21.20 0.000 215.2924 259.7286

The Beta coefficient of spdlaw turns out to be positive. This shows that on average increase in speed
limit from 55 to 65 resulted in a 64.66 increase in accidents.

However the above regression is very simple and needs to be refined by incorporating omitted
variables. Also other checks have to be performed to check robustness of the model.

Time series data generally follows seasonal trend so it needs to be de-seasonalized. To do this we
introduce dummy variables of seasonal months keeping January as base group. The purpose of doing
this is that according to our experience road traffic does not remain uniform across all months. There
might be abnormal traffic in holiday seasons or vacations. This means that we have accounted for
seasonality and it will not reflect in our Beta coefficient of spdlaw. Thus as expected it decreases to
55.75 from 64.66.

6|Page
Speed limit impact on accidents in U.S 2016

Source SS df MS Number of obs = 107


F( 13, 93) = 52.79
Model 610764.841 13 46981.9108 Prob > F = 0.0000
Residual 82767.7763 93 889.976089 R-squared = 0.8807
Adj R-squared = 0.8640
Total 693532.617 106 6542.76054 Root MSE = 29.832

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

spdlaw 55.75342 10.34311 5.39 0.000 35.21405 76.2928


t 1.274385 .151619 8.41 0.000 .9732995 1.57547
feb -11.05216 14.06398 -0.79 0.434 -38.98043 16.87611
mar 47.45123 14.06643 3.37 0.001 19.51809 75.38437
apr 41.84351 14.07051 2.97 0.004 13.90226 69.78476
may 68.70764 14.08425 4.88 0.000 40.73911 96.67616
jun 81.21103 14.08187 5.77 0.000 53.24722 109.1748
jul 119.3811 14.08113 8.48 0.000 91.41875 147.3434
aug 132.5511 14.08203 9.41 0.000 104.587 160.5153
sep 76.72121 14.08455 5.45 0.000 48.75209 104.6903
oct 36.44682 14.0887 2.59 0.011 8.469449 64.4242
nov 54.17244 14.09449 3.84 0.000 26.18358 82.1613
dec 61.95437 14.50684 4.27 0.000 33.14667 90.76207
_cons 179.1655 11.51249 15.56 0.000 156.304 202.027

High coefficients for June, July and Aug might be because people travel more during these seasons
which increases the chance of road accidents.

To check if we have not incorporated any important explanatory variable, we perform Omitted variable
bias test:

Ramsey RESET test using powers of the fitted values of rtotacc


Ho: model has no omitted variables
F(3, 90) = 4.09
Prob > F = 0.0091

We reject null hypothesis and our model requires more explanatory variables. Keeping this in mind we
introduce Belt law variable to our model and see its impact.

7|Page
Speed limit impact on accidents in U.S 2016
Source SS df MS Number of obs = 107
F( 14, 92) = 53.21
Model 617292.975 14 44092.3554 Prob > F = 0.0000
Residual 76239.6414 92 828.691754 R-squared = 0.8901
Adj R-squared = 0.8733
Total 693532.617 106 6542.76054 Root MSE = 28.787

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

spdlaw 51.4665 10.09684 5.10 0.000 31.4133 71.5197


t .8817183 .2024304 4.36 0.000 .479674 1.283763
feb -10.6595 13.57183 -0.79 0.434 -37.61433 16.29534
mar 48.23656 13.57636 3.55 0.001 21.27273 75.20039
apr 43.02151 13.58391 3.17 0.002 16.0427 70.00032
may 70.75463 13.61023 5.20 0.000 43.72354 97.78572
jun 83.65069 13.61616 6.14 0.000 56.60782 110.6935
jul 122.2134 13.62509 8.97 0.000 95.15281 149.274
aug 135.7761 13.63702 9.96 0.000 108.6918 162.8604
sep 80.33886 13.65195 5.88 0.000 53.22492 107.4528
oct 40.45715 13.66985 2.96 0.004 13.30764 67.60665
nov 58.57543 13.69073 4.28 0.000 31.38446 85.7664
dec 66.23753 14.08139 4.70 0.000 38.27067 94.20438
beltlaw 31.69068 11.29103 2.81 0.006 9.265722 54.11563
_cons 185.2741 11.32022 16.37 0.000 162.7911 207.757

Belt law was introduced in U.S on January 1986. Beta coefficient of Beltlaw is significant but
interestingly it is positive. Introduction of beltlaw increased accidents by 31.69 on average. This might
be due to psychology of the driver who after feeling safe with a seat belt drives rashly and causes
accidents. There have been studies conducted which also support the claim that seat belts increase road
accidents.

Ramsey RESET test using powers of the fitted values of rtotacc


Ho: model has no omitted variables
F(3, 88) = 0.62
Prob > F = 0.6060

Our Omitted variables score has also improved but still there are some more variables that need to be
incorporated. If we introduce unemployment rate variable then:

8|Page
Speed limit impact on accidents in U.S 2016
Source SS df MS Number of obs = 107
F( 15, 91) = 49.88
Model 618325.958 15 41221.7305 Prob > F = 0.0000
Residual 75206.6587 91 826.446799 R-squared = 0.8916
Adj R-squared = 0.8737
Total 693532.617 106 6542.76054 Root MSE = 28.748

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

spdlaw 49.18807 10.28705 4.78 0.000 28.75411 69.62202


t .8138291 .2110793 3.86 0.000 .394546 1.233112
feb -10.35066 13.55625 -0.76 0.447 -37.27849 16.57717
mar 47.20202 13.5895 3.47 0.001 20.20814 74.1959
apr 40.95338 13.69104 2.99 0.004 13.75781 68.14895
may 68.1126 13.79569 4.94 0.000 40.70914 95.51605
jun 81.93707 13.68382 5.99 0.000 54.75585 109.1183
jul 122.2543 13.60667 8.98 0.000 95.22634 149.2823
aug 134.1295 13.69796 9.79 0.000 106.9202 161.3388
sep 77.86513 13.81183 5.64 0.000 50.42963 105.3006
oct 38.12014 13.81044 2.76 0.007 10.6874 65.55288
nov 56.6161 13.78404 4.11 0.000 29.23581 83.9964
dec 63.88831 14.21843 4.49 0.000 35.64514 92.13147
beltlaw 28.55584 11.61913 2.46 0.016 5.475858 51.63582
unem -3.097905 2.770952 -1.12 0.267 -8.602061 2.406251
_cons 214.836 28.75721 7.47 0.000 157.7133 271.9586

P-value of unem variable lies in the insignificant region so it does not have much of an impact on rural
accidents.

Hence we can drop unemployment rate variable. Our final model has Speed law, beltlaw, Time (t) and
seasonal dummies explaining Total accidents on rural roads while unemployment is an insignificant
variable. Our model is explained 89.16% with R-Square: 0.8916

After adding explanatory independent variables we need to check if multi-collinearity exists between
them. For this we make use of Variance Inflation factors (VIF):

Variable VIF 1/VIF

t 5.50 0.181715
beltlaw 4.31 0.232275
unem 3.13 0.319326
spdlaw 2.82 0.354684
sep 1.90 0.525567
oct 1.90 0.525672
may 1.90 0.526797
nov 1.90 0.527688
aug 1.87 0.534341
apr 1.87 0.534881
jun 1.87 0.535446
jul 1.85 0.541535
mar 1.84 0.542904
feb 1.83 0.545570
dec 1.81 0.552294

Mean VIF 2.42

9|Page
Speed limit impact on accidents in U.S 2016

Our maximum VIF value is 5.5 while our Mean VIF is 2.42 which is well below the bench mark of 5. So
there is no multi-collinearity in our independent variables.

Since our data is time series so we use tsset t, monthly command and then use archlm command.

. tsset t, monthly
time variable: t, 1960m2 to 1969m1
delta: 1 month

. estat archlm
LM test for autoregressive conditional heteroskedasticity (ARCH)

lags(p) chi2 df Prob > chi2

1 0.168 1 0.6817

H0: no ARCH effects vs. H1: ARCH(p) disturbance

Our P-value comes out to be 0.6817 so we fail to reject our null hypothesis. There is no Autoregressive
conditional heteroscadasticity. Hence our data is Homoscedastic.

Result and Final Regression Equation:


From the above analysis we come to a conclusion that there is a significant impact of speed law on
number of road accidents. Hence we fail to accept our null hypothesis. Our regression equation is:

(Y)rtotacc=51.46spdlaw+31.69beltlaw+0.88t-10.66Feb+48.24mar+43.02apr+70.75may+
83.65jun+122.21jul+135.77aug+80.34sep+40.45oct+58.57nov+66.23dec

10 | P a g e
Speed limit impact on accidents in U.S 2016

Method No. 2
A second way of running the regression is first stabilizing the series that follow time trend and then
running the regression. Stabalization is done by first differencing. Lets begin by seeing which series is
following a time trend. To check whether the data is stationary or not, we used E-views and plotted the
data points.

As we can see from the above graph that our datas intercept is not at zero while the slope has an
upward trend, which shows that this variable is non-stationary. This means that we will not be able to
run simple regression on our current data, hence we will have to make some adjustments to it.

We applied the same process on other variable Unem (unemployment) and found that it was also non-
stationary.

11 | P a g e
Speed limit impact on accidents in U.S 2016

UNEM
12

11

10

4
10 20 30 40 50 60 70 80 90 100

The graph for unemployment showed similar result since the graph was downward sloping.
Furthermore, we conducted Augmented Dickey-Fuller Unit Root Test on rtotacc and Unem to check the
stationarity of the data set variables. We found that the p-value of rtotacc and Unem to be 0.99 and
0.3479 respectively, which proved that the data is non-stationary.

Unit Root Test For RTOTACC


Null Hypothesis: RTOTACC has a unit root
Exogenous: None
Lag Length: 11 (Automatic - based on SIC, maxlag=12)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic 3.560903 0.9999


Test critical values: 1% level -2.589273
5% level -1.944211
10% level -1.614532

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation


Dependent Variable: D(RTOTACC)

12 | P a g e
Speed limit impact on accidents in U.S 2016

Method: Least Squares


Date: 04/13/16 Time: 17:48
Sample (adjusted): 13 108
Included observations: 96 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

RTOTACC(-1) 0.044944 0.012621 3.560903 0.0006


D(RTOTACC(-1)) -0.700190 0.106793 -6.556529 0.0000
D(RTOTACC(-2)) -0.688927 0.116919 -5.892339 0.0000
D(RTOTACC(-3)) -0.568434 0.126405 -4.496939 0.0000
D(RTOTACC(-4)) -0.512617 0.124617 -4.113549 0.0001
D(RTOTACC(-5)) -0.519907 0.108218 -4.804243 0.0000
D(RTOTACC(-6)) -0.727022 0.094408 -7.700892 0.0000
D(RTOTACC(-7)) -0.782306 0.106961 -7.313969 0.0000
D(RTOTACC(-8)) -0.592724 0.124746 -4.751457 0.0000
D(RTOTACC(-9)) -0.466485 0.128481 -3.630772 0.0005
D(RTOTACC(-10)) -0.529481 0.119947 -4.414281 0.0000
D(RTOTACC(-11)) -0.381094 0.110628 -3.444822 0.0009

R-squared 0.544640 Mean dependent var 2.593750


Adjusted R-squared 0.485010 S.D. dependent var 50.65582
S.E. of regression 36.35205 Akaike info criterion 10.14085
Sum squared resid 111003.6 Schwarz criterion 10.46139
Log likelihood -474.7607 Hannan-Quinn criter. 10.27042
Durbin-Watson stat 1.975286

Unit Root Test For UNEM

Null Hypothesis: UNEM has a unit root


Exogenous: None
Lag Length: 0 (Automatic - based on SIC, maxlag=12)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -0.844485 0.3479


Test critical values: 1% level -2.586753
5% level -1.943853
10% level -1.614749

13 | P a g e
Speed limit impact on accidents in U.S 2016

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation


Dependent Variable: D(UNEM)
Method: Least Squares
Date: 04/13/16 Time: 17:48
Sample (adjusted): 2 108
Included observations: 107 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

-
UNEM(-1) 0.006216 0.007361 -0.844485 0.4003

R-squared 0.004070 Mean dependent var -0.028972


Adjusted R-squared 0.004070 S.D. dependent var 0.567518
S.E. of regression 0.566362 Akaike info criterion 1.710137
Sum squared resid 34.00124 Schwarz criterion 1.735117
-
Log likelihood 90.49233 Hannan-Quinn criter. 1.720263
Durbin-Watson stat 2.209081

To make the above data points (total number of rural accidents and unemployment) stationary, we took
first difference Augmented Dickey-Fuller Unit Root Test on on rtotacc and Unem on first difference.
From the first difference, we found that the p-value of both rtotacc and Unem to be 0.0000 which
showed that the data has now become stationary.

14 | P a g e
Speed limit impact on accidents in U.S 2016

Unit Root Test For RTOTACC with first difference

Null Hypothesis: D(RTOTACC) has a unit root


Exogenous: Constant, Linear Trend
Lag Length: 10 (Automatic - based on SIC, maxlag=12)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -8.553854 0.0000


Test critical values: 1% level -4.056461
5% level -3.457301
10% level -3.154562

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation


Dependent Variable: D(RTOTACC,2)
Method: Least Squares
Date: 04/13/16 Time: 17:41
Sample (adjusted): 13 108
Included observations: 96 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

D(RTOTACC(-1)) -7.513557 0.878383 -8.553854 0.0000


D(RTOTACC(-1),2) 5.833317 0.816606 7.143372 0.0000
D(RTOTACC(-2),2) 5.158729 0.740416 6.967339 0.0000
D(RTOTACC(-3),2) 4.591235 0.648253 7.082469 0.0000
D(RTOTACC(-4),2) 4.071850 0.554643 7.341382 0.0000
D(RTOTACC(-5),2) 3.548651 0.483845 7.334269 0.0000
D(RTOTACC(-6),2) 2.824484 0.433957 6.508670 0.0000
D(RTOTACC(-7),2) 2.037037 0.371384 5.484996 0.0000
D(RTOTACC(-8),2) 1.430969 0.288756 4.955634 0.0000
D(RTOTACC(-9),2) 0.943708 0.196003 4.814752 0.0000
D(RTOTACC(-10),2) 0.398673 0.109659 3.635573 0.0005
C -3.494836 8.720760 -0.400749 0.6896
@TREND(1) 0.299753 0.137352 2.182368 0.0319

R-squared 0.804624 Mean dependent var 0.895833

15 | P a g e
Speed limit impact on accidents in U.S 2016

Adjusted R-squared 0.776377 S.D. dependent var 75.92856


S.E. of regression 35.90573 Akaike info criterion 10.12500
Sum squared resid 107005.4 Schwarz criterion 10.47225
Log likelihood -472.9998 Hannan-Quinn criter. 10.26536
F-statistic 28.48513 Durbin-Watson stat 1.996966
Prob(F-statistic) 0.000000

1st Diff(rtot )
200

100

-100

-200

-300

-400

-500
10 20 30 40 50 60 70 80 90 100

16 | P a g e
Speed limit impact on accidents in U.S 2016

Unit Root Test For Unem with first difference

Null Hypothesis: D(UNEM) has a unit root


Exogenous: Constant, Linear Trend
Lag Length: 1 (Automatic - based on SIC, maxlag=12)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -9.054714 0.0000


Test critical values: 1% level -4.047795
5% level -3.453179
10% level -3.152153

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation


Dependent Variable: D(UNEM,2)
Method: Least Squares
Date: 04/13/16 Time: 17:42
Sample (adjusted): 4 108
Included observations: 105 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

-
D(UNEM(-1)) 1.312866 0.144993 -9.054714 0.0000
D(UNEM(-1),2) 0.184486 0.097427 1.893586 0.0611
C 0.048384 0.114075 0.424140 0.6724
-
@TREND(1) 0.001504 0.001821 -0.825965 0.4108

R-squared 0.572244 Mean dependent var 0.004762


Adjusted R-squared 0.559538 S.D. dependent var 0.849704
S.E. of regression 0.563925 Akaike info criterion 1.729561
Sum squared resid 32.11920 Schwarz criterion 1.830664
-
Log likelihood 86.80196 Hannan-Quinn criter. 1.770530
F-statistic 45.03859 Durbin-Watson stat 2.051122

17 | P a g e
Speed limit impact on accidents in U.S 2016

Prob(F-statistic) 0.000000

This shows that by applying the first difference our data has become stationary.

1st diff(unemp)
2

-1

-2

-3

-4

-5
10 20 30 40 50 60 70 80 90 100

Now we are ready to run the run the regression, on the data.

18 | P a g e
Speed limit impact on accidents in U.S 2016

Source SS df MS Number of obs = 107


F( 15, 91) = 7.87
Model 146255.567 15 9750.37112 Prob > F = 0.0000
Residual 112796.657 91 1239.52371 R-squared = 0.5646
Adj R-squared = 0.4928
Total 259052.224 106 2443.88891 Root MSE = 35.207

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

stdiffunemp -10.14038 8.453573 -1.20 0.233 -26.93237 6.6516


spdlaw .4245773 10.99296 0.04 0.969 -21.41159 22.26074
beltlaw 3.270204 9.992475 0.33 0.744 -16.57862 23.11903
wkends -2.237807 3.90837 -0.57 0.568 -10.0013 5.52569
feb 60.41767 18.11391 3.34 0.001 24.43662 96.39872
mar -.1929481 17.05431 -0.01 0.991 -34.06924 33.68334
apr 38.65245 17.06046 2.27 0.026 4.76394 72.54097
may 25.03917 16.76973 1.49 0.139 -8.271851 58.35018
jun 51.66942 17.70881 2.92 0.004 16.49304 86.84579
jul 17.14284 17.47816 0.98 0.329 -17.57538 51.86106
aug -49.04038 16.91911 -2.90 0.005 -82.64812 -15.43265
sep -31.82191 16.99697 -1.87 0.064 -65.5843 1.94047
oct 28.45865 16.6692 1.71 0.091 -4.652658 61.56997
nov 21.48958 17.03561 1.26 0.210 -12.34957 55.32874
dec -51.67342 18.57374 -2.78 0.007 -88.56789 -14.77896
_cons 20.0465 54.80308 0.37 0.715 -88.81309 128.9061

As we can see from the above table, our Co-efficient of Determination (R-squared) for the above
variables is 0.56, which shows that our data is significantly fitted to the regression line i.e.56% of our
response variable has been explained by the model. For our model, we see that spdlaw has p-value of
0.97, which shows that it is insignificant. Stdiffunemp and beltlaw have a p-value of 0.23 and 0.74
respectively i.e. they do not have a significant impact on the above model.

We conduct Variance Inflation Factor (VIF) to check for multicollinearity.


Variable VIF 1/VIF

feb 2.18 0.458296


spdlaw 2.15 0.465837
beltlaw 2.12 0.471024
jun 2.09 0.479503
dec 2.06 0.485416
jul 2.03 0.492242
stdiffunemp 1.97 0.508052
apr 1.94 0.516640
mar 1.93 0.517013
nov 1.93 0.518149
sep 1.92 0.520508
aug 1.90 0.525309
may 1.87 0.534709
oct 1.85 0.541179
wkends 1.30 0.767959

Mean VIF 1.95

Our maximum VIF is 2.18 while Mean VIF is 1.95. This shows that our data has no multi-collinearity.

19 | P a g e
Speed limit impact on accidents in U.S 2016

We also conduct Omitted variable test to see if any significant variable is missing from our model. We
fail to reject null hypothesis as P-value is greater than 0.05(5% level of significance). Hence there is no
Omitted variable bias in our model.

Ramsey RESET test using powers of the fitted values of stdiffrtot


Ho: model has no omitted variables
F(3, 88) = 0.16
Prob > F = 0.9203

Result and Final Regression Equation:

From the above analysis we come to a conclusion that there is no significant impact of speed law on
number of road accidents. Hence we accept our null hypothesis. Our regression equation is:

(Y)stdiffrtot=-10.14stdiffunemp+0.42spdlaw+3.27beltlaw-2.24wkends+60.42Feb-
0.19mar+38.65apr+25.04may+51.67jun+17.14jul-49.04aug-31.82sep+28.46oct+21.49nov-
51.67dec+20.04

20 | P a g e
Speed limit impact on accidents in U.S 2016

Conclusion
In this paper we presented two ways to progress with regression analysis. The purpose was to discuss
the two methods. Two statisticians can adopt two different methods and approach at two different
results. It is improper to pass judgment on any method and as such passing judgment falls out of the
domain of our paper. One method yielded results that showed significant impact of road accidents
while the other method did not find significance of speed law on number of accidents.

However, we must take the results presented with a pinch of salt since there might be other variables
like weather conditions or road quality which have an impact on road accidents and could not be
included due to unavailability of data.

21 | P a g e
Speed limit impact on accidents in U.S 2016

Bibliography
1. https://books.google.com.pk/books?id=64vt5TDBNLwC&pg=PA375&lpg=PA375&dq=econometr
ic+wooldridge+traffic+2&source=bl&ots=Jz-
WnHgzub&sig=5uIzW8BA2uE0RSGTroxYTj3wIw8&hl=en&sa=X&ved=0ahUKEwjHopP0xY3MAhU
GYJoKHVTZDr8Q6AEIJjAC#v=onepage&q=econometric%20wooldridge%20traffic%202&f=false
(Link for Book and Question)
2. http://ajbuckeconbikesail.net/Econ3503/Data/TRAFFIC2.txt (Link for data set)

http://ajbuckeconbikesail.net/Econ3503/Autocorr_Hwk/Autocorrelation_Key.html

22 | P a g e

Вам также может понравиться