NUST Business School

ECO-215 Fundamentals of Econometrics

Final Project

Submitted to:
Dr. Mohsin Ali

Submitted by:
Mahnoor Sajid 416678
Areeba Salman 414138
Ayesha Sajid 413486
Iesha Malik 411384

Arisha Tahir 429538

BBA 2K22(A)

Date: 15th May,2024

TABLE OF CONTENTS
 Does time served in prison (tottime) or average sentence length (avgsen) influence the
likelihood of future arrests (nfarr86)?

Abstract:
This paper examines the correlation between multifactorial determiners and future arrest
probabilities in the criminal justice system. It uses an econometric analysis to determine the
statistical relationships between critical indicators and future arrest probabilities. The report
will also examine whether time served in prison (tottime) or average sentence length (avgsen)
influence the likelihood of future arrests (nfarr86) or not.
The dependent variables include total time spent in prison, average sentence length,
percentage of previous crimes, and recent unemployment duration. Demographic
characteristics like race/ethnicity are also considered. The model reveals key trends: longer
sentences lead to fewer convictions for offenders, and longer joblessness slightly increases
arrests. Records indicate a higher likelihood of good behaviour, while racial minorities may
be involved in arrest odds. The report offers action strategies that combine educational and
vocational programmes to prevent incarceration with rehabilitation. Sentencing options like
community service or drug treatment centres may be more effective. Investment in education
and community building is also necessary to address systemic issues. So, although time
served in prison and average sentence length do impact the likelihood of future arrests, this
impact is not statistically significant, when we analyse the econometric model. Other
econometric concepts are also tested via the software STATA in this paper. A corss-sectional
dataset was used, acquired from Wooldridge datasets: Wooldridge Source: R.C. Fair (1978),
namely CRIME1.

Introduction:
Econometrics has become a valuable tool for assessing the complex links between factors and
future criminal activities. It helps identify factors influencing future arrests, which is crucial
for criminal policy-making and crime prevention strategies. Econometric analysis helps
researchers discover causal relationships and subtle dependencies, enabling evidence-centred
policy choices. Additional research is being conducted to understand the association between
relevant variables and future arrest probability.
Factors considered include total prison time, average sentence length, prior convictions, and
inmate race or ethnicity, particularly black or Hispanic inmates. Previous studies have
provided a layered picture of the vicissitudes leading to these variables, but this research
provides essential data on the connection between imprisonment occurrence and future arrest
chances, which is often used for evidence-based laws and criminal justice systems.
Literature Review:
In the article ‘Estimating effects of short-term imprisonment on crime using random judge
assignments’, authors use random assignment of judges as an instrumental variable to assess
whether incarceration can deter crime as well as to study the impacts of imprisonment on
recidivism rates. The theoretical framework is built upon the deterrence theory which
suggests that when people feel threatened by punishments, they are more likely to fear
committing or being associated with any criminal activity. In contrast to previous studies
which proposed that short term imprisonment may have a criminogenic effect increasing the
frequency of arrests in the future, authors conducting this study, mitigate methodological
issues such as endogeneity and include focus on external cofounding factors which may
result in the findings being bias, to propose the positive influence short term imprisonment
has on crime (Wermink et al., 2023).
In addition, Enes Al Weswasi, Fredrik Sivertsson, Olof Bäckman & Anders Nilsson also
contribute to the understanding of how different types of sentencing practices impact the
likelihood of individuals with a criminal record reoffending. The concept of incapacitation
is also integrated with authors suggesting that prolonged sentences remove criminals from the
society for a significant period thus leading to a reduction in crime rates and decreasing their
opportunity to decrease further offences. However, the article concludes that policy reforms
being imposed in Sweden proposing shorter length sentences, were declared not entirely due
to their association with recidivism rates, rather due to the fact that there are several other
factors including quality of rehabilitation programs and support systems surrounding the
convict which play a pivotal role in reducing criminal activities (Al Weswasi et al., 2022)

Furthermore, Gerald Gaes and Julia Laskorunsky draw upon theories based on criminology
and economics to provide us with a deeper insight into the existing complexities between
these variables. Coupled with deterrence theory, selective incapacitation theory is brought
into play as authors state that by targeting a specific segment entailing high risk offenders and
giving them longer sentences, maximum benefit can be attained while also minimising
pressure on prison resources. The main finding of this article however reiterates that it is not
always long-term incarcerations that lead to reduced crime rates, in fact short term
imprisonment along with factors such as prison release practices, parole guidelines etc also
lead to a positive effect on criminal activities among the population (The Relationship
between Sentence Length, Time Served, and State Prison Population Levels - Task Force on
Long Sentences, n.d.) These sources collectively validate the model's assumptions and
methodological approach. So, this model was formed after careful review of past studies and
existing literature relevant to the topic., as described below:
Economic/Theoretical Model:
The economic model is:
nfarr86 = β0 + β1tottime + β2avgsen + β3pcnv+ β4durat + β5black + β6hispan + u
Where nfarr86 refers to the likelihood of arrests in the future. Β0 serves as the intercept. β1,
β2, β3, β4, β5 and β6 are the coefficients to be estimated, tottime refers to the total time spent
by convicts in prison, avgsen means the average sentence length, durat is the duration of
recent unemployment, black and hispanic are the dummy variables respectively referring to
being black or hispanic and u is the error term.
The problem is adequately formulated as the question gives a comprehensive insight into
the impact of length of imprisonment on crime rates. Formed on the basis of both deterrence
and incapacitation theories, the problem is designed within a detailed theoretical framework
that takes into consideration the complexity of crime dynamics and the multifaceted effects of
imprisonment, allowing the causal relationship between the two to be estimated more
accurately.
This model is also supported by economic theory as the model is backed by several
economic theories surrounding criminology. The deterrence theory suggests that when a
threat of punishment exists, likelihood of crimes being committed significantly decreases.
This generates the hypothesis that imprisonment can deter future criminal activities.
Moreover, the incapacitation theory puts forward the idea that when offenders are removed
from society, their ability to commit crimes is hampered, thus causing a reduction in crime
rates. This theory also proposes that maximum social benefits are attained when longer
sentences are given. This led to the inclusion of variables such as time served and sentence
length in our model. Management theories also play a role in substantiating the model as
theories related to rehabilitation and a decrease in recidivism are important to fully
understand the importance of taking underlying contributors to criminal behaviour also into
account.
Moreover, this model is also supported by literature as elaborated in literature review.
To some extent, our model is based on intuition and endorses one’s logical expectations
regarding how imprisonment might influence crime rates. For example, it is a common
assumption that longer sentences may deter crimes and removing offenders from society can
optimise use of prison resources while also proving to be greatly beneficial. However, our
model essentially requires a lot of empirical data to theorise these claims which can only be
done by a rigorous econometrics analysis.
All the independent variables chosen successfully pass through the common-sense filter,
theory filter as well as the scholarly literature filter. Total time spent in prison was chosen
based on the general assumption that the longer a person is in prison, higher would be the
chances of future arrests due to convicts becoming institutionalised and not having sufficient
opportunities for rehabilitation. This can be supported by the deterrence theory discovered in
our literature review, particularly through the work of Wermink et al. (2023) which supports
the relation between time spent in prison and criminal activities. Similarly average sentence
length was added to the model as common sense suggests that the length and severity of the
punishment plays a significant role in threatening individuals from reoffending.
Incapacitation theory lies at the core of this concept and it is endorsed by the work of Gaes
and Laskorunsky (Task Force on Long Sentences, n.d.) who cover the relationship between
sentence length and crime rates. Furthermore, proportion of prior conviction was proposed
and added as individuals with a history of prior convictions will be under increased
supervision and be receiving more support thus reducing their likelihood of future arrests.
This is also supported by incapacitation theory which suggests that targeting the segment
entailing high risk offenders with a criminal record will lead to a vast number of social
benefits. Al Weswasi et al. (2022) provides sufficient research strengthening the relationship
between prior convictions and future behaviour. Lastly, duration of recent employment is
often considered to be a reason behind criminal activities as offenders tend to be motivated by
financial necessity. This is also supported by various economic theories suggesting that
financial strains can be driving factors for criminal behaviour (Agnew, 2015). Gaes and
Laskorunsky (Task Force on Long Sentences, n.d.) also substantiated the claim by shedding
light on the relation between unemployment and recidivism. In addition, including
demographic factors like black and hispanic was integral for our model as racial and ethnic
minorities, due to systematic injustice and inequality, may be at high risk for imprisonment.
This can be supported by various studies done on racial discrimination and the subsequent
social problems it causes all over the world. (Ulmer et al., 2014)

Econometric model:
nfarr86 = β0 + β1tottime + β2avgsen + β3pcnv+ β4durat + β5black + β6hispan + u
The model is formulated in the general functional form as the variable nfarr86 depends on
several factors like tottime, avgsen, pcnnv, durat, black and hispan. Each parameter has a
coefficient estimated by regression, and the model contains an intercept term β0 and error
term u.
This model is based on a cross-sectional dataset acquired by Wooldridge Source: R.C. Fair
(1978), namely CRIME1. The percentage of likelihood of future arrests (nfarr86) is
statistically dependent on several key variables, including the total number of hours spent in
prison(tottime), the average sentence length in months(avgsen), the percentage of
proportions of prior convictions(pcnv), and the duration of recent unemployment in
months(durat). Being black and Hispanic also causes variations in this likelihood, according
to this model. By using a thorough econometric analysis, it is found that changes in these
variables have a significant impact on the likelihood of future arrests. The regression model
shows that each unit change in total time served in prison, average sentence length,
percentage of prior convictions, and length of recent unemployment has a unique effect on
the likelihood of future arrests, along with demographic factors like being black or Hispanic.
  Regression:

The estimated model, through regression, would be:

nfarr86 = 0.1729 + 0.00475tottime – 0.00583avgsen – 0.0756pcnv+ 0.0117durat +

0.2606black + 0.08663hispan + u

Using STATA, we can regress the model in the following way:

[ Note that: Log file is attached, comprising of each command]

Through regression, we can see the strength of the relationships between the regressors and
the regressand. It can also tell us about the goodness of fit, and the interpretations of each
parameter, as well as whether the parameter has a statistically significant impact on the
regressor or not.
Goodness of Fit:
The coefficient of determination (R²) shows how well the independent variables explains the
variation in the dependent variable. An R² of 4.14% suggests that approximately 4.14% of the
variation in nfarr86 is explained by the variation of independent variables included in the
model. This relatively low R² indicates a poor fit, meaning the model does not capture a
significant portion of the variability in nfarr86.
Model Specification:
The model is specified as follows:
𝑛𝑓𝑎𝑟𝑟86=𝛽0+𝛽1tottime+𝛽2avgsen+𝛽3penv+𝛽4durat+𝛽5black+𝛽6hispa
nn
This structure is typical for an econometric regression model, where nfarr86 is the dependent
variable and tottime, avgsen, penv, durat, black, and hispan are the independent variables.
 1. Explaining Poor Fit:
As the model is not a good fit, we have assumed there are several reasons for
it, as stated below:
 Omitted Variables: There could be other important variables that
influence nfarr86 which are not included in the dataset. For example,
socioeconomic factors, educational attainment, employment status,
neighborhood characteristics, etc. (narr86 -# times arrested, ptime86-
mos. in prison during 1986, qemp86-# quarters employed, inc86-legal
income etc). These variables can significantly impact criminal
behavior and should be incorporated into the model to improve its
explanatory power. After running a regression by including narr86, the
R-value rose to 65.53%. This suggests that this variable is crucial for
explaining felony arrests. [check log file]
 Measurement Error: The variables included might have measurement
errors or might not accurately capture the concepts they are intended to
measure, from the data set.
Even though, as our model is strongly supported by theory, we realize that we need to be
cautious while providing implications because of the existence of Heteroskedasticity, and a
low-R square value.

Interpretations:
We can interpret all the parameters in the model by this in the following way:
nfarr86 is a dependent variable, which depends on the other parameters.
β0 is an intercept. Its value is 0.1729 which means that when all the other parameters are zero
then the likelihood of future arrests is 0.1729%.
β1 is the coefficient of tottime, with the value of 0.00475. This coefficient is positive, so
when the total number of hours spent in prison increases by 1 hour, then the likelihood of
future arrests increases by 0.00475%, keeping the rest of the factors constant.
β2 is the coefficient of avgsen, with the value of 0.0058. This coefficient is negative, and
implies that longer average sentence lengths lead to a decrease in the likelihood of future
arrests. This aligns with the common expectation that harsher sentences act as a prevention to
criminal behaver. When the average sentence length increases by a month, then likelihood of
future arrests decreases by 0.0058%, keeping other factors constant.
β3 is the coefficient of pcnv, with the value of 0.0856. This coefficient is also negative,
indicating that a higher proportion of prior convictions is associated with a decrease in the
likelihood of future arrests, this seems unlikely in real-world scenario as repeat offenders tend
to get arrested more often, as per Stolzenberg, L., et. al (2020, June 27). In this model, 1%
increase in proportion of prior convictions causes a 0.0856% decrease in likelihood of getting
arrested in the future, keeping the rest of the factors accounted for.
β4 is the coefficient for durat, with the value of 0.0117. This coefficient is positive, showing
that longer durations of recent unemployment are associated with a slight increase in the
likelihood of future arrests. Keeping all other factors constant, when duration of
unemployment recently increases by one month, then the likelihood of future arrests
increases by 0.011%.
β5 is the coefficient for the dummy variable black, with the value 0.2606. This means that
when all other factors are accounted for, then black people are 0.26% more likely to get
arrested in the future than nonblacks. Similarly, β6 is the coefficient for the dummy variable
hispan, with the value of 0.08663. In this model, when all other parameters are accounted for,
then Hispanic people are 0.086% more likely to get arrested in the future than non-Hispanic
people.

Estimating only using Ordinary Least Square or OLS initially may possibly cause
problems such as heteroskedasticity, multicollinearity, etc. To check if our model is free of
these issues or not:
Post-Estimation Analysis:

 Testing multicollinearity:

For this, we test the correlation between the variables via STATA, as shown in log-file. [ Note
that: Log file is attached, comprising of each command]

Although perfect multicollinearity does not exist in the model, by STATA, we can tell that
tottime and avgsen are closely correlated as their value is 0.9.
To fix multicollinearity, remove one variable. Here, there is some correlation between avgsen
and tottime as by STATA, we can see its correlation value is 0.9 which is ultimately close to
1. So, by removing tottime or avgsen, this model can be fixed and free of multicollinearity.
We are fixing this by removing avgsen. So, the new model would be

nfarr86 = β0 + β1tottime + β3pcnv+ β4durat + β5black + β6hispan + u

Estimate this model through regression again via STATA.

nfarr86 = 0.1728 + 0.00063tottime – 0.0756pcnv+ 0.0117durat + 0.2606black +
0.08663hispan

Now, we run the correlation command again on STATA to see that multicollinearity is fixed,
there is no value that closely correlated, with values close to 1. We can also check vif to see
that multicollinearity is fixed. For this, refer to the log-file. [Note that: Log file is attached, comprising of
each command]

In the research question, we asked if time served in prison (tottime) or average sentence
length (avgsen) influence the likelihood of future arrests (nfarr86)? Both of them do have an
impact, however they are correlated so it may be better to remove one variable to fix this
multicollinearity. To see if tottime and avgsen have a significant impact on the likelihood of
future arrests, we test it by first checking the heteroskedasticity.
Multicollinearity can also be checked by checking vif of the parameters before and after
removing avgsen. On STATA, by checking this we can see that before removing avgsen, there
was multicollinearity as some vif values were more than 3. However, after removing avgsen,
centered vif is less than 3 for all the variables.
[Mentioned in log-file]

 Testing heteroskedasticity:
[Note that: Log file is attached, comprising of each command]

For this, we conduct Breusch Pagan Test:

H₀= the model does not have heteroskedasticity
H1= the model has heteroskedasticity
First errors were predicted, then a variable for the square of errors was generated. After that,
the original model was regressed by the square of errors instead of the dependent variable. By
this, we can see that there is heteroskedasticity as the p-value 0.0000 is less than the level of
significance 0.05. This is statistically significant and so we reject H ₀, heteroskedasticity
exists in the model. Refer to the log-file for the commands.
As there is heteroskedasticity, to see if the variables of this model are significant, we use
robust standard errors to regress the original model. [As shown in log-file via STATA]
By this, we can now do t-tests to answer the econometric question
 Testing: ‘Does time served in prison (tottime) or average sentence length (avgsen)
influence the likelihood of future arrests (nfarr86)?’
For tottime:
H₀= the time served in prison does not significantly impact the likelihood of future arrests
H1= the time served in prison significantly impacts the likelihood of future arrests
Alpha value is 5%
T-value is calculated in STATA by dividing the coefficient of tottime with the robust standard
error. This value is 0.6. The critical value is 1.96.
T-value<Critical value
So, we fail to reject the null. The time served in prison does not significantly impact the
likelihood of future arrests. This impact is not statistically significant.

For avgsen:
H₀= the average sentence length does not significantly impact the likelihood of future arrests
H1= the average sentence length significantly impacts the likelihood of future arrests
Alpha value is 5%
T-value is calculated in STATA by dividing the coefficient of tottime with the robust standard
error. This value is -0.64. The critical value is -1.96.
T-value<Critical value
So, we fail to reject the null. The average sentence length does not significantly impact the
likelihood of future arrests. This impact is not statistically significant.

We can conclude that time served in prison (tottime) or average sentence length (avgsen)
may influence the likelihood of future arrests (nfarr86), however this impact is not
statistically significant as per this model, even though the model is supported by theories and
literature.

Dummy variables

Descriptive Statistics:

Policy Implications:
 Total Time Spent in Prison (tottime):
Findings: According to research, an increase in the total sentence spent in prison can be
slightly connected to the ones who are likely to come arrested.
Policy Implication: The policy designs should focus on the conflicting issue of incarceration’s
prevention along with rehabilitation intended. This could entail the introduction, within the
correctional facilities, programs that help with education, job retraining, mental health
services that aid proper reintegration into society after the release.
 Average Sentence Length (avgsen):
Findings: A longer mean sentence length does entails a decrease in statistically significant
arrests.
Policy Implication: Stronger sentence regimes for habitual offenders are perhaps what will
thwart crime. Nevertheless, extreme sums may not completely solve the problem. Adopting
non-custodial penalties style, e.g., community service or drug rehabilitation programs, can be
successful intervention mechanisms for an offender’s reform.
  Proportion of Prior Convictions (pcnv):
Findings: A randomized pattern of high conviction proportions resulted in a onefold drop in
future arrest probabilities.
Policy Implication: People with prior convicts likely are falling behind due to the gap in
social support system, vocational training, and fair employment, they may benefit from innate
interventions like individualized supervision, access to job opportunities, and community
programs that help in reducing crime rate.
 Duration of Recent Unemployment (durat):
Findings: More extended periods of unemployment of last year are linked to this with a
marginal increase in the risk of detainment in the upcoming future.
Policy Implication: Ensuring jobs by organizing job training programs, vocational education,
and re-integration support can reduce the possibility of folks who turn to crime because of the
economic hardship. Strategies aim at distributing available funds and providing assistance for
the population to find permanent work that enables them to get out of poverty.
 Demographic Factors (black and hispan):
Findings: If a Black or Native Americans are related to higher proportion of arrests in future.
Policy Implication: Merging systemic incompatibilities and obstacles faced by Black and
Hispanic populations is essential. A policy should invest in schooling, job training and
community development programs so as to try to deal with the origin facets of criminal
justice system that impel people to join it.

Conclusion:
References:
 Wooldridge Source: R.C. Fair (1978), “A Theory of Extramarital Affairs,” Journal of Political
Economy 86, 45-61, 1978 http://fmwww.bc.edu/ec-p/data/wooldridge/datasets.list.html
 Wermink, H. T., Blokland, A. A. J., Been, J., Schuyt, P. M., Tollenaar, N., & Apel, R.
(2023). Estimating effects of short-term imprisonment on crime using random judge
assignments. Justice Quarterly, 1–27. https://doi.org/10.1080/07418825.2023.2193618
 Al Weswasi, E., Sivertsson, F., Bäckman, O., & Nilsson, A. (2022). Does sentence
length affect the risk for criminal recidivism? A quasi-experimental study of three
policy reforms in Sweden. Journal of Experimental Criminology.
https://doi.org/10.1007/s11292-022-09513-1
 The Relationship between Sentence Length, Time Served, and State Prison Population
Levels - Task Force on Long Sentences. (n.d.). Counciloncj.foleon.com.
https://counciloncj.foleon.com/tfls/long-sentences-by-the-numbers/the-relationship-
between-sentence-length-time-served-and-state-prison-population-levels
 Stolzenberg, L., D’Alessio, S. J., & Flexon, J. L. (2020, June 27). The Usual Suspects: Prior
Criminal Record and the Probability of Arrest. Retrieved from https://orcid.org/0000-0002-
1655-5348 https://doi.org/10.1177/1098611120937304
 Ulmer, J., Painter-Davis, N., & Tinik, L. (2014). Disproportional Imprisonment of Black and
Hispanic Males: Sentencing Discretion, Processing Outcomes, and Policy Structures. Justice
Quarterly, 33(4), 642–681. https://doi.org/10.1080/07418825.2014.958186
 Agnew, R. (2015). Strain, Economic Status, and Crime. The Handbook of Criminological
Theory, 209–229. https://doi.org/10.1002/9781118512449.ch11