PAGE 1
University of Florida Economics of Cybercrime: Demographic and Socioeconomic Determinants of Cybercrime Victimology Megan Wolf Thesis Advisor: Dr. Michelle Phillips 2 8 Mar . 202 2
PAGE 2
Wolf 1 I. INTRODUCTION Cybercrime refers to a broad category of criminal offenses that involve computers or are perpetrated through computer networks . In recent years, t he Internet has become an essential component of conducting business and connecting individual s. Unfortunately, the migration from paper to digital records and increased online transactions create new targets and vulnerabilities for criminal agents to exploit. Over the past five years, the ime Complaint Center (IC3) has received over 2 million complaints and estimates over $13.3 billion in reported total losses by individual victims of cybercrime . In order to give context to the significance of socioeconomic and demographic determinants of cybercrime victimology, it is essential to understand the underlying social engineering that often contributes to cybercriminal offenses. Cybercrime always involv es some form of computing, but it is still necessary to examine the human elements of such crimes. Social engineering refers to the exploi tation of human vulnerabilitie s through tactics of manipulation, er a computer system, or to steal personal . A socioeconomic background likely ha s an impact on his psychological interactions with the world both physically and virtually , and these interaction patterns are exploited by hackers and other cybercriminal offenders. The goal of this paper is to examine the demographic and socioeconomic factors under which cybercrime victimization occurs and postulate which factors are correlated wit h cybercriminal offenses against individuals of a population.
PAGE 3
Wolf 2 II. SAMPLE The sample used for this study is the 50 states of the United States of America in 201 3 2019 . The study leverages demographic and socioeconomic differences among states in order to estimate the impact of each independent variable on cybercrime victimology. III. DEPENDENT VARIABLES The dependent variables for this study are the total cybercrime victim count and victim financial loss in U.S. d ollars for each U.S. state in the 7 observed years from 2013 2019 . Annualized cybercrime victim counts and losses are obtained through c omplainant self reported location to the IC3 . Due to the self reported nature of the complaints, under reporting is a possible concern with some crime types. All complaints are reviewed by IC3 analysts to categorize offenses by crime type(s) and adjust loss amounts based on available data . The dependent variable represents an aggregate total of the crime types reported including phishing/vishing/smishing/pharming, non payment/non delivery, extortion, personal data breach, and identity theft . See Appendix Definitions for a more detailed list of crime categorizations and how the IC3 defines each of the crime types. The dependent variables of cybercrime victim count and financial loss are adjusted for state population (victimization per 100,000 people) and examined in separate regressions .
PAGE 4
Wolf 3 IV. INDEPENDENT VARIABLES TABLE 1: INDEPENDENT VARIABLE DESCRIPTIONS Independent Variable Description Source Census Table Terrestrial Crime (non cybercrime) 1 Violent Crimes Violent crime offenses per 100,000 people, yearly FBI UCR N/A Property Crimes Property crime offenses per 100,000 people, yearly FBI UCR N/A Internet Penetration related Broadband Rate Ratio of broadband subscriptions to overall Internet subscriptions, percent, 1 year ACS B28002 Internet Subscription Households with an Internet subscription per 100,000 people , percent, 1 year ACS B28002 Demographic Factors Percent Non White Non white population to total population, percent, 1 year ACS US Census B02001 Percent Elderly Percentage of population age 65 or older, percent, 1 year ACS US Census B28005 Percent Elderly with Computing Device Ratio of population 65 years and over with a computing device to total population age 65 or older, percent, 1 year ACS US Census B28005 Percent Speak Non English Language at Home Percentage of population that speaks a language other than English at Home year ACS US Census B16001 Socioeconomic Factors Median Income Median household annual income, 1 year ACS US Census B19013 College Educated higher, percent, 1 year ACS US Census B15003 Poverty Rate Percentage of population below the poverty level, percent, 1 year ACS US Census B17001 Unemployment Rate Percentage of unemployed individuals out of the employable population, percent, 1 year ACS US Censu s B23025 US Census data comes from 1 year and 5 year American Community Survey tables as indicated above (American Community Survey 5 Year Data and American Community Survey 1 Year Data). through the Uniform Crime Reporting Program (Crime Data Explorer). Broadband Rate and Internet Subscription Internet connectivity is required for the perpetration of many types of cybercriminal acts. The Internet subscription variable is represented by the number of households with any kind of Internet subscription per 100,000 people in the , respectively. Because a greater online presence creates more potential points of vulnerability and attack, a greater number of households with Internet subscriptions is expected to be associated with a higher rate of cybercrime victimization. Furthermore, examining the ratio of broadband connections to overall 1 This paper builds on the independent variables ( like broadband rate violent crimes and property crimes) used in a for which the data covers an earlier period of 2004 2010 and the dependent variable focuses on perpetrator identity r ather than victim socioeconomic and demographic composition (Park, et al. 2).
PAGE 5
Wolf 4 infrastructure. It is unclear whether the adoption of broadband is expected to result in higher rates of cybercrime victimizati the speeds at which victims are connected. Violent Crimes and Property Crimes The data for vi olent crimes and property crimes Uniform Crime Report (UCR). For each state, the UCR categorizes violent crimes as offenses including homicide, rape, robbery, and aggravated assault. The UCR classifies offenses like arso n, burglary, larceny theft, and motor vehicle theft as property crimes ( . Although many cybercrimes do not follow similar spatial rules as traditional crimes due to their online nature, it is still important to consider these crime variables because it is expected that states with greater occurrences of terrestrial criminal offenses would have higher instances of cybercriminal offenses. Some crimes have both a terrestrial and cyber componen t (e.g., theft of physical mail to obtain personal information to be used in identity theft schemes online). Moreover, including violent crimes and property crimes as variables can help with understanding how cybercrime may relate to the traditional econom ic theory of crime. ( Weber ) . Socioeconomic cybercrime refers to financially motivated computer and Internet crime (Park et al. 5). As such, c ybercriminal offenses that may be considered crimes of impulse like crimes against children would most likely be correlated with violent criminal offenses whereas crimes of wealth like ransomware would
PAGE 6
Wolf 5 most likely be correlated with terrestrial property cr ime. Overall, increased terrestrial crime is expected to be correlated with increased cybercrime. Non white Ratio populations, and [all] People of Color (BIPOC) are more lik ely to have their identities stolen than White people (21 percent compared to 15 percent), and BIPOC people are the least likely to avoid any financial impact due to cybercrime (47 percent compared to 59 percent of all . It is expected that a greater non white population ratio in a state will be correlated with higher victim counts . Percent Elderly and Percent Elderly with a Computing Device Several previous studies have recognized the importance of age as a factor of cybercrime breakdowns by age groups: 10,724 victims under age 20 suffered a total loss of $421,169,232; 44,496 victims age 20 29 suffered a total loss of $174,673,470; 52,820 victims age 30 39 suffered a total loss of $332,208,189; 51,864 victims age 40 49 suffered a total loss of $529,231,267; 50,608 victims age 50 59 suffered a total loss of $589,624,844; 68,013 victims fall victim to some type of financial fraud or internet sc 28% of the total fraud losses were sustained by victims over the age of 60, resulting in Therefore, it is expected that states with a higher perce ntage elderly (people 65 years of age or older) will have higher rates of victimization . Because the elderly population typically experiences substantially
PAGE 7
Wolf 6 higher victim counts and losses, the percent elderly in a population is utilized to represent the re lationship of age to victimization rather than the median age. Additionally, because there is likely a trend of extremely limited to no computing or internet usage for elders past a certain age, the percent elderly with a computing device is accounted for as an additional independent variable. Elderly people are especially susceptible to internet fraud; however, i t is also possible that beyond a certain age, internet utilization for personal purposes decreases due to technological aversion or dependent liv ing situations. Vic timization may be less likely to occur in situations where computer usage is eliminated. As such, it is expected that a higher ratio of elderly people with a computing device will be correlated with higher rates of cybercrime victimizati on than percent of elderly people alone. Speak Language Other than English at Home may help explain why speaking a language other than English could contribute to higher rates of cybercrime victimization. In this type of email scam, a perpetrator tricks the victim into sending money under the guise of needing it to escape a foreign country while promising the fortune of a prince. These types of scam e mails and phishing attempts often contain many spelling mistakes and grammatical errors. However, a ccording to ils are a strategic method that scammers use to Individuals who do not sp eak English at home may be less likely to identify incorrect spelling and grammatical errors. It is expected that states with a higher percentage of people that speak a language other than English at home will have higher rates of cybercrime victimization . This
PAGE 8
Wolf 7 variable is represented by the ratio o f non English speakers at home who speak English less than year American Community Survey for all listed non English languages 2 . Percent College E ducated Because hackers and scammers employ strategic methods like spelling and grammatical errors to select gullible targets, a higher education rate is likely to deter many common cybercriminal schemes. This study uses the percentage of the state populat ion over 25 years old any higher level of education including graduate degrees and professional school in order to determine the education rate. It is expected that states with a higher education rate will have lower rates of cybercrime victimization. Median Income It is expected that states with a higher median income will have lower victim count ; however, i ndividuals with higher incomes potentially have more to lose if successfully targeted by cybercriminal attacks. This may incentivize perpetrators to target higher income individuals. However, higher incomes may also afford better technological safeguards like antivirus software and newer devices with updated software and higher security standards . For example, people who cannot afford to upgrade to a new operating system may be more vulnerable to exploits (e.g., Windows 7 stopped receiving updates in 2020 and may be more vulnerable to attack than Windows 10). 2 The data for non English speakers at home does not include any of the native languages of North America (e.g., Navajo) due to limited state by state data availability.
PAGE 9
Wolf 8 Poverty Rate It is expected that states with a higher poverty rate will have lower victim counts and losses. Cybercriminals are less likely to benefit from targets that live below the poverty line because such victims yield little to no financial benefits. Unemploymen t Rate It is expected that states with a higher unemployment rate will have higher rates of victimization . The psychological and financial impact of unemployment may cause an individual to accept fraudulent opportunities they may not otherwise consider wit h stable employment , however; without steady income , financial losses may be more limited . V. SU M MARY STATISTICS TABLE 2: SUMMARY STATISTICS Variable N Mean Std. Dev. Min Max ln(Victim Count) 350 4.33 .326 3.673 5.892 ln(Victim Loss) 350 12.543 .702 11.057 14.633 ln(Violent Crime) 350 5.826 .391 4.631 6.793 ln(Property Crime) 350 7.757 .265 7.074 8.276 ln(Internet) 350 10.315 .096 9.971 10.503 Broadband Rate 350 65.463 39.682 4.562 99.871 ln(Median Income) 350 10.956 .174 10.544 11.371 Percent Elderly 350 15.558 2.016 8.999 21.128 Percent Elderly with Computing Device 350 78.406 7.225 56.146 92.923 Unemployment Rate 350 5.698 1.69 2.593 10.96 Percent Non White 350 23.293 12.687 5.024 75.901 Percent Non English Speaking 350 36.948 5.861 17.84 49.444 Poverty Rate 350 13.662 3.141 7.27 24.049 Percent College Educated 350 20.635 3.812 13.152 31.654 The dependent variables were adjusted to account for population differences between states and represent the cybercrime victim count (or victim loss) per 100,000 people in respective population. The Internet subscriptions variable was similarly adjusted to represent the household Internet subscriptions per 100,000 people in e V iolent crime per 100,000 people and property crime per 100,000 people were already adjusted the in the FBI UCR dataset to account for population differences between states. Additionally,
PAGE 10
Wolf 9 before running the regressions, n ominal variables (e.g., Internet subscriptions) were log transformed because their distributions are skewed. An extended version of the summary of statistics including non log adjusted variables can be found in Table 1A of the Appendix. VI. REGRESSIONS AND RESULTS The matrix of correlations between the independent variables can be seen in Table 2A of the Appendix . The regressions discussed in this paper utilize log transformed variables. This study holds that significant correlation is present at or above | 0.7|. V ariables with a correlation greater than | 0.7| include poverty rate and median income ( 0.847), poverty rate and percent college educated ( 0.72 3 ), and median income and percent college educated (+0.8 11 ). I n attempt to account for multicollinearity, regressions excluding each of these highly correlated indepe ndent variables individually (one at a time) were run. Table 3 shows the results for the cybercrime victim count OLS regressions. The results for cybercrime victim loss OLS regressions, cybercrime victim count state year fixed effects regressions, and cybercrime victim loss state year fixed effects regressions can be found in Table 4 A, Table 5 A, and Table 6 A of the Appendix, respectively. Terrestrial Crime O LS Regression Results Violent crime was significant at the 1% significance level for all regressions for victim count. The regressions suggest that a 1% increase in violent crime victim count is associated with approximately a 0.19% 0.21% increase in cy bercrime victim count, which supports the hypothesized relationship. Contrary to the hypothesized relationship, the property crime victim count was not statistically significant in any of the cybercrime victim count regressions. A deeper review of the rela ted criminological concepts is likely required to better understand the complex relationship between cyber/virtual and terrestrial crime.
PAGE 11
Wolf 10 TABLE 3 : CYBERCRIME VICTIM COUNT OLS REGRESSIONS ln(Victim Count) (1) (2) (3) ( 4 ) ln(Violent Crime) 0.2061 *** (5.64) 0.2120 *** (5.86) 0.1854 *** (5.17) 0.1918 *** (5.22) ln(Property Crime) 0.0880 ( 1.48) 0.0438 ( 0.68) 0.0143 ( 0.23) 0.0232 ( 0.36) ln(Internet) 0.3348 (1.25) 0.1315 (0.51) 0.3588 (1.37) 0.3190 (1.20) Broadband Rate 0.0006 ( 1.29) 0.0004 ( 0.95) 0.0007 ( 1.63) 0.0006 ( 1.35) % Elderly 0.0102 ( 1.46) 0.0061 ( 0.78) 0.0014 (0.17) 0.0005 (0.06) % Elderly w/ Computers 0.0314 *** (12.46) 0.0296 *** (10.61) 0.0279 *** (9.95) 0.0281 *** (9.97) Unemployment Rate 0.0724 *** (6.98) 0.0654 *** (6.35) 0.0664 *** (6.96) 0.0696 *** (6.73) % Non White 0.0044 *** (3.33) 0.0025 (1.57) 0.0016 (1.06) 0.0020 (1.27) % Non English 0.0064 *** ( 3.09) 0.0061 *** ( 2.94) 0.0057 *** ( 2.75) 0.0058 *** ( 2.79) Poverty Rate 0.0281 *** ( 3.78) 0.0150 ( 1.43) 0.0087 ( 0.81) % College Educated 0.0066 ( 1.38) 0.0165 *** ( 2.86) 0.0154 *** ( 2.60) ln(Median Income) 0.2531 (1.17) 0.8181 *** (4.48) 0.6620 ** (2.49) Constant 1.6644 ( 0.61) 2.8911 ( 0.88) 11.3277 *** ( 3.80) 9.1183 ** ( 2.26) N 350.0000 350.0000 350.0000 350.0000 R 2 0.6558 0.6553 0.6614 0.6620 F Statistic 58.5445 *** 58.4042 *** 60.0147 *** 55.0121 *** Residual Sum of Squares 12.8031 12.8232 12.5956 12.5712 Rank 12.0000 12.0000 12.0000 13.0000 RMSE 0.1946 0.1948 0.1930 0.1931 Adjusted R 2 0.6446 0.6440 0.6504 0.6500 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
PAGE 12
Wolf 11 Internet Penetration Regression Results Contrary to the hypothesized relationships, Internet subscriptions and broadband rate are not statistically significant in any of the cybercrime victim count regressions. This may be because there is little variability between state Internet infrastructure in recent years. 3 Demographic Factors OLS Regression Results Percent non white is statistically significant at the 1% significance level for regression (1) and implies that a one percentage point rise in the non white population is associated with a 0.44% increase in cybercrime victim count . The coefficient for regression (1) was positive and supports the hypothesized relationship . Percent elderly was not statistically significant in any of the regressions for cybercrime victim count. These unexpected results may be reconciled by the outcomes observed in percent elderly with a computing device. Percent elderly with a computing device was statistically significant in all regressions for victim count at the 1% si gnificance level . A one percentage point rise in elderly with a computing device implies a 2.79% 3.14% increase in cybercrime victim count . Speaking a non English language at home was statistically significant at the 1% significance level for all of the cy bercrime victim count regressions; however, the sign of the coefficients was unexpectedly negative. A one percentage point rise in percent non English language speakers is associated with 0.57% 0.64% decrease in cybercrime victim count . 3 Internet penetration variables were found to be sign ificant in a previous study by Park et al.; however, this study utilized state data during 2004 2010 when there was more variability in Internet infrastructure between states. Additionally, due to the nature of hacking, computational power and network spee d are likely greater determinants for perpetration than victimization when a substantial amount of the population has access to computing devices and Internet connection as is the case for 2013 2019.
PAGE 13
Wolf 12 Socioeconomic Factors OLS Regression Results Median income was statistically significant in regression (3) at the 1% significance level and regression (4) at the 5% significance level. The regressions for which median income was statistically significant imply that a 1% increase in median income is associated with a 0.66% 0.82% increase in cybercrime victim count ; however , this does not support the hypothesized relationship for victim count. T he benefits and utilization of higher cost security measures may have been over estimated, or it is possible t hat extra security measures do little to dissuade perpetrators who are incentivized by the possibility of higher payouts. Percent college educated was statistically significant in reg ressions (3) and (4) at the 1% significance level and imply that a one percentage point increase in the percent college educated is associated with a 1.54% 1.65% decrease in cybercrime victim count . This result supports the hypothesis that a higher rate of education is correlated with fewer cybercrime victims. Poverty rate was statistically significant at the 1% significance level in regression (1) for victim count , suggesting that a one percentage point increase in the poverty rate implies a 2.81% decrease in cybercrime victim count. This is consistent with the hypothesized relationship. Unemployment rate was statistically significant at the 1% significance level for all cybercrime victim count regressions. The positive coefficient signs support the hypothesized relationship that victim counts would be greater for higher unemployment rates . The regressions impl y that a one percentage point increase in the unemployment rate is associated with a 6. 54 % 7.24% increase in cybercrime victim count .
PAGE 14
Wolf 13 VII. CONCLUSION This study indicates that there is a statistically significant impact of socioeconomic and demographic factors on cybercrime victimology . Again, it is important to note that due to the self reported nature of the complaint data, under reporting, especially among certain groups (e.g., minorities and the elderly), may explain many of the findings. Therefore, it would be worthwhile to re examine the extent to which different socioeconomic determinants impact cybercrime victimization given a more extensive and comprehensive data source in the future. Regarding the statistical methods used for analysis in this paper, t he above analysis utilized OLS regressions; however , utilizing fixed effects (state and year) as shown in the Appendix may provide a better empirical model for the longitudinal panel data. Moreover, it may be beneficial to utilize a Weighted Least Squares approach to account for heteroskedasticity in the dataset. Similarly, p revious literature suggests employing a fixed effects regression with Driscoll and Kraay standard errors to account for state speci fic omitted variables as well as interstate correlations in residuals ; however, because the current dataset has N=50 and T= 7, a larger T (more years) may be necessary to leverage the robust benefits of Driscoll Kra a y errors . Furthermore, it would be of interest to examine specific crime categorizations beyond total victim counts and losses separately . A wide range of offenses classified as cybercriminal acts were included in the totals studied in this paper . D ifferent classifications of crime have wide ranging motive s and required technological expertise. As such, it can be expected that because the cr iminological fundamentals of the specific crime types vary , so too will the demographic and socioeconomic qualities of the victims. A more granular approach to cybercrime offense classifications could provide more insight as to the socioeconomic and demogr aphic qualities that may impact victimization .
PAGE 15
Wolf 14 Works Cited Crime Data Explorer , Federal Bureau of Investigation, https://crime data explorer.app.cloud.gov/explorer/national/united states/prelim quarter. Justia , Justia , 15 July 2021, https://www.justia.com/criminal/offenses/other crimes/cybercrimes/#:~:text=Other%20Offenses%20Cybercrimes ,Cybercrimes,goals%20other%20than%20financial%20gain. Malwarebytes , Savanta Inc., 2021, https:// malwarebytes main dev.cphostaccess.com/resources/2021 demographics of cybercrime report/index.html. ACM Transactions on Management Information Systems , vol. 10, no. 4, 2019, pp. 1 23., https://doi.org/10 .1145/3351159. Carnegie Mellon University Information Security Office , Carnegie Mellon University, https://www.cmu.edu/iso/aware/dont take the bait/social engineering.html. Joseph Steinberg: CyberSecurity, Privacy, & Artificial Intelligence (AI) Advisor , 2 Sept. 2019, https://josephsteinberg.com/why scammers make spelling and grammar mistakes/. United States, FBI, Internet Crime Complaint Center IC3. 2019 Intern et Crime Report , IC3. https://www.ic3.gov/Media/PDF/AnnualReport/2019_IC3Report.pdf. Accessed 1 Oct. 2021. United States, FBI, Internet Crime Complaint Center IC3. 2020 Elder Fraud Report , IC3. https://www.ic3.gov/Media/PDF/AnnualReport/2020_IC3ElderFraud Report.pdf. Accessed 1 Oct. 2021.
PAGE 16
Wolf 15 United States, FBI, Internet Crime Complaint Center IC3. 2020 Internet Crime Report , IC3. https://www.ic3.gov/Media/PDF/AnnualReport/2020_IC3Report.pdf. Accessed 1 Oct. 2021. United States, U.S. Census Bureau, American C ommunity Survey 1 Year Data (2005 2020) . https://www.census.gov/data/developers/data sets/acs 1year.html . Accessed 1 Dec . 2021. United States, U.S. Census Bureau, American Community Survey 5 Year Data (2009 2019) . https://www.census.gov/data/developers/da ta sets/acs 5 year.html . Accessed 1 Dec . 2021. Regional Science and Urban Economics , vol. 48, 29 Apr. 2014, pp. 1 11., https://doi.org/10.1016/j.regsciurbeco.2014.04.006.
PAGE 17
Wolf 16 VIII. APPENDIX TABLE 1A: SUMMARY STATISTICS EXTENDED Variable N Mean Std. Dev. Min Max Victim Count 350 80.839 35.147 39.382 362.112 ln(Victim Count) 350 4.33 .326 3.673 5.892 Victim Loss 350 364677.59 302418.16 63393.051 2264190.2 ln(Victim Loss) 350 12.543 .702 11.057 14.633 Violent Crimes 350 365.057 141.76 102.6 891.7 ln(Violent Crime) 350 5.826 .391 4.631 6.793 Property Crimes 350 2417.152 610.315 1180.6 3928.7 ln(Property Crime) 350 7.757 .265 7.074 8.276 Internet Subscription 350 30308.288 2836.032 21403.166 36429.075 ln(Internet) 350 10.315 .096 9.971 10.503 Broadband Rate 350 65.463 39.682 4.562 99.871 Median Income 350 58161.417 10316.44 37963 86738 ln(Median Income) 350 10.956 .174 10.544 11.371 Percent Elderly 350 15.558 2.016 8.999 21.128 Percent Elderly with Computing Devices 350 78.406 7.225 56.146 92.923 Unemployment Rate 350 5.698 1.69 2.593 10.96 Percent Non White 350 23.293 12.687 5.024 75.901 Percent Non English Speaking at Home 350 36.948 5.861 17.84 49.444 Poverty 350 13.662 3.141 7.27 24.049 Percent College Educated 350 20.635 3.812 13.152 31.654 TABLE 2A : MATRIX OF CORRELATIONS Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (1) ln(Violent Crime) 1.000 (2) ln(Property Crime) 0.579 1.000 (3) ln(Internet) 0.309 0.570 1.000 (4) Broadband Rate 0.048 0.228 0.581 1.000 (5) ln(Median Income) 0.205 0.407 0.513 0.291 1.000 (6) % Elderly 0.190 0.349 0.469 0.357 0.092 1.000 (7) % Elderly w/ Computers 0.012 0.206 0.606 0.637 0.680 0.176 1.000 (8) Unemployment Rate 0.317 0.417 0.683 0.545 0.422 0.263 0.506 1.000 (9) % Non White 0.379 0.360 0.399 0.017 0.219 0.165 0.087 0.298 1.000 (10) % Non English 0.128 0.184 0.302 0.152 0.015 0.196 0.161 0.238 0.456 1.000 (11) Poverty Rate 0.390 0.504 0.654 0.255 0.871 0.020 0.597 0.616 0.115 0.142 1.000 (12) % College Educated 0.337 0.479 0.570 0.185 0.811 0.148 0.496 0.318 0.055 0.044 0.723 1.000 Significant correlation is present TABLE 3 A: MATRIX OF CORRELATIONS NON LOG AJDUSTED Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (1) Violent Crime 1.000 (2) Property Crime 0.576 1.000 (3) Internet 0.304 0.577 1.000 (4) Broadband Rate 0.064 0.217 0.587 1.000 (5) Median Income 0.183 0.390 0.475 0.287 1.000 (6) % Elderly 0.169 0.326 0.477 0.357 0.092 1.000 (7) % Elderly w/ Computers 0.025 0.196 0.604 0.637 0.660 0.176 1.000 (8) Unemployment Rate 0.316 0.428 0.685 0.545 0.392 0.263 0.506 1.000 (9) % Non White 0.347 0.368 0.401 0.017 0.252 0.165 0.087 0.298 1.000 (10) % Non English 0.004 0.167 0.308 0.152 0.006 0.196 0.161 0.238 0.456 1.000 (11) Poverty Rate 0.367 0.518 0.650 0.255 0.847 0.020 0.597 0.616 0.115 0.142 1.000 (12) % College Educated 0.330 0.461 0.570 0.185 0.805 0.148 0.496 0.318 0.055 0.044 0.723 1.000 Significant correlation is present
PAGE 18
Wolf 17 TABLE 4 A : CYBERCRIME VICTIM LOSS OLS REGRESSIONS ln(Victim Loss ) ( 5 ) ( 6 ) ( 7 ) ( 8 ) ln(Violent Crime) 0.3436 *** (4.42) 0.3176 *** (4.28) 0.3954 *** (4.99) 0.2764 *** (3.67) ln(Property Crime) 0.6591 *** ( 5.21) 0.3979 *** ( 3.01) 0.5239 *** ( 3.73) 0.3560 *** ( 2.69) ln(Internet) 1.6957 *** (2.97) 1.2405 ** (2.34) 0.8740 (1.51) 1.6214 *** (2.96) Broadband Rate 0.0004 ( 0.41) 0.0001 ( 0.15) 0.0013 (1.38) 0.0005 ( 0.56) % Elderly 0.0247 * ( 1.66) 0.0119 (0.74) 0.0086 (0.48) 0.0253 (1.52) % Elderly w/ Computers 0.0673 *** (12.55) 0.0551 *** (9.62) 0.0575 *** (9.29) 0.0521 *** (9.00) Unemployment Rate 0.0489 ** ( 2.22) 0.0704 *** ( 3.33) 0.0009 ( 0.04) 0.0618 *** ( 2.91) % Non White 0.0096 *** (3.42) 0.0004 ( 0.11) 0.0065 * (1.94) 0.0014 ( 0.42) % Non English 0.0084 * (1.91) 0.0106 ** (2.49) 0.0096 ** (2.11) 0.0114 *** (2.67) Poverty Rate 0.0724 *** (4.58) 0.1502 *** (6.96) 0.1631 *** (7.42) % College Educated 0.0096 (0.94) 0.0107 ( 0.84) 0.0313 ** ( 2.57) ln(Median Income) 2.2650 *** (5.12) 0.1622 (0.40) 3.0960 *** (5.68) Constant 8.1509 ( 1.40) 30.3575 *** ( 4.51) 1.4992 ( 0.23) 43.0125 *** ( 5.19) N 350.0000 350.0000 350.0000 350.0000 R 2 0.6628 0.6862 0.6420 0.6922 F Statistic 60.3845 *** 67.1906 *** 55.1057 *** 63.1693 *** Residual Sum of Squares 58.0021 53.9704 61.5692 52.9294 Rank 12.0000 12.0000 12.0000 13.0000 RMSE 0.4143 0.3996 0.4268 0.3963 Adjusted R 2 0.6518 0.6760 0.6304 0.6813 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
PAGE 19
Wolf 18 TABLE 5 A: CYBERCRIME VICTIM COUNT FIXED EFFECTS REGRESSIONS ln(Victim Count) ( 9 ) ( 10 ) ( 11 ) ( 12 ) ln(Violent Crime) 0.1097 (0.90) 0.1080 (0.89) 0.1040 (0.86) 0.1204 (0.99) ln(Property Crime) 0.2300 * ( 1.73) 0.1790 ( 1.36) 0.1967 ( 1.49) 0.1881 ( 1.42) ln(Internet) 0.4243 ( 0.80) 0.2162 ( 0.41) 0.3555 ( 0.68) 0.2640 ( 0.50) Broadband Rate 0.0005 (0.21) 0.0002 ( 0.10) 0.0007 (0.28) 0.0001 (0.04) % Elderly 0.1125 *** ( 3.08) 0.0956 *** ( 2.60) 0.0860 ** ( 2.36) 0.0934 ** ( 2.54) % Elderly w/ Computer 0.0085 ( 0.99) 0.0085 ( 1.01) 0.0088 ( 1.05) 0.0094 ( 1.12) Unemployment Rate 0.0347 ** ( 1.98) 0.0337 * ( 1.94) 0.0258 ( 1.58) 0.0341 * ( 1.96) % Non White 0.0061 (0.43) 0.0019 (0.14) 0.0054 (0.38) 0.0034 (0.24) % Non English Speaking 0.0239 *** (2.74) 0.0254 *** (2.94) 0.0248 *** (2.87) 0.0251 *** (2.90) Poverty Rate 0.0056 ( 0.38) 0.0251 (1.37) 0.0254 (1.38) % College Educated 0.0041 ( 0.16) 0.0284 ( 1.07) 0.0288 ( 1.09) ln(Median Income) 1.3292 ** (2.50) 1.0560 ** (2.37) 1.5393 *** (2.72) Constant 11.2741 ** (2.07) 6.4194 ( 0.74) 1.1900 ( 0.16) 7.6527 ( 0.87) N 350.0000 350.0000 350.0000 350.0000 R 2 0.6120 0.6203 0.6194 0.6219 F Statistic 26.2568 *** 27.2005 *** 27.0880 *** 25.7713 *** Residual Sum of Squares 4.1438 4.0547 4.0651 4.0378 Rank 18.0000 18.0000 18.0000 19.0000 RMSE 0.1210 0.1197 0.1199 0.1197 Adjusted R 2 0.5215 0.5318 0.5306 0.5321 0.9113 0.9031 0.9060 0.9100 u 0.3878 0.3653 0.3722 0.3804 e 0.1210 0.1197 0.1199 0.1197 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
PAGE 20
Wolf 19 TABLE 6 A: CYBERCRIME VICTIM LOSS FIXED EFFECTS REGRESSIONS ln(Victim Loss ) ( 13 ) ( 14 ) ( 15 ) ( 16 ) ln(Violent Crime) 0.2771 (0.83) 0.2860 (0.86) 0.2721 (0.82) 0.2703 (0.81) ln(Property Crime) 0.0893 (0.25) 0.0509 (0.14) 0.0634 (0.17) 0.0625 (0.17) ln(Internet) 1.1272 (0.78) 0.9639 (0.67) 1.0347 (0.72) 1.0245 (0.71) Broadband Rate 0.0061 (0.90) 0.0068 (1.01) 0.0063 (0.94) 0.0064 (0.94) % Elderly 0.1382 ( 1.39) 0.1476 ( 1.46) 0.1513 ( 1.51) 0.1504 ( 1.48) % Elderly w/ Computer 0.0330 ( 1.42) 0.0335 ( 1.45) 0.0324 ( 1.40) 0.0324 ( 1.39) Unemployment Rate 0.0215 ( 0.45) 0.0224 ( 0.47) 0.0228 ( 0.51) 0.0219 ( 0.46) % Non White 0.0250 (0.65) 0.0286 (0.74) 0.0265 (0.69) 0.0268 (0.69) % Non English Speaking 0.0174 (0.73) 0.0163 (0.68) 0.0167 (0.70) 0.0167 (0.70) Poverty Rate 0.0170 (0.43) 0.0025 ( 0.05) 0.0028 ( 0.06) % College Educated 0.0207 (0.30) 0.0365 (0.50) 0.0365 (0.50) ln(Median Income) 0.7189 ( 0.49) 0.9319 ( 0.76) 0.9853 ( 0.63) Constant 0.4825 (0.03) 11.0335 (0.46) 11.8833 (0.58) 12.5972 (0.52) N 350.0000 350.0000 350.0000 350.0000 R 2 0.7700 0.7701 0.7703 0.7703 F Statistic 55.7242 *** 55.7625 *** 55.8258 *** 52.5388 *** Residual Sum of Squares 30.7665 30.7503 30.7234 30.7231 Rank 18.0000 18.0000 18.0000 19.0000 RMSE 0.3297 0.3296 0.3295 0.3301 Adjusted R 2 0.7163 0.7165 0.7167 0.7157 0.7244 0.7752 0.7357 0.7350 u 0.5346 0.6122 0.5497 0.5496 e 0.3297 0.3296 0.3295 0.3301 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
PAGE 21
Wolf 20 Definitions Definitions of crime types as designated by IC3 ( : Confidence/Romance Fraud: An individual believes they are in a relationship (family, friendly, or romantic) and are tricked into sending money, personal and financial information, or items of value to the perpetrator or to launder money or items to assist the perpetrator. This inc ludes the Crimes Against Children: Anything related to the exploitation of children, including child abuse. Extortion: Unlawful extraction of money or p roperty through intimidation or undue exercise of authority. It may include threats of physical harm, criminal prosecution, or public exposure. Identity Theft: Someone steals and uses personal identifying information, like a name or Social Security number , without permission to commit fraud or other crimes and/or (Account Takeover) a fraudster obtains account information to perpetrate fraud on existing accounts. Malware/Scareware/Virus: Software or code intended to damage, disable, or capable of copying i tself onto a computer and/or computer systems to have a detrimental effect or destroy data. Non Payment/Non Delivery: In non payment situations, goods and services are shipped, but payment is never rendered. In non delivery situations, payment is sent, but goods and services are never received. Personal Data Breach: A leak/spill of personal data which is released from a secure location to or confidential data is copied, transmitted, viewed, stolen or used by an unauthorized individual.
PAGE 22
Wolf 21 Phishing/Vishing/Smishing/Pharming: To use unsolicited email, text messages, and telephone calls purportedly from a legitimate company requesting personal, fina ncial, and/or login credentials. Ransomware: A type of malicious software designed to block access to a computer system until money is paid. Spoofing: Contact information (phone number, email, and website) is deliberately falsified to mislead and appear to be from a legitimate source. For example, spoofed phone numbers making mass robo calls; spoofed emails sending mass spam; forged websites used to mislead and gather personal information. Often used in connection with other crimes.
PAGE 23
Wolf 22 Other Variable Considerations The following independent variables were previously included in the proposal paper but are no longer considered in the regressions: population, percent of households with one or more computing devices, median age, Gini index, percent employed in computer/engineering/science occupations, percent females employed in computer/engineering/science occupations, a dummy for Net Neutrality laws, and a dummy for state level data privacy laws. Population is now accounted for in the indep endent variable by adjusting the victim counts and losses by the state population per 100,000 people. Computing devices had a correlation coefficient greater than 0.9 with the broadband rate. Because computing devices and broadband rate are effectively ser ving as the same proxy for the factor of Internet penetration infrastructure, only the broadband rate is used in the regressions. Moreover, due to concerns of collinearity, median age is no longer used in favor of only using percent elderly as the age fact or in the regression. Because the physical separation between perpetrators and victims is not restricted to state borders, it does not make sense to include it as an indicator of victimization. We would have to look at perpetrator locations for Gini Index to be an effective measure as seen in Park et al. Regarding both employment in computer/engineering/science occupations variables, after examining previous literature, it was decided that these variables would be more relevant if perpetration was studied r ather than victimization. For the dummy variables previously considered, state level data privacy laws are likely still an important factor; however, this variable is unlikely to impact the sample used in this study because although some laws were passed w ithin the given timeframe of 2017 2019, none of the laws took effect until after 2020. It is important to acknowledge that K 12 computer science education could be an additional variable to consider; however, it is beyond the scope of this study to standa rdize what
PAGE 24
Wolf 23 constitutes a K 12 computer science education curriculum in a state. Moreover, it is likely too early to see an impact from recently established K 12 computer science programs on the other studied variables. Another variable that is not include d but is relevant is social media usage. Unfortunately, there were no public datasets that break down social media usage by state. Other studies have established that there are differences in the ways males and females are impacted by cybercrime. In the c ase of this study, limited variation between state gender ratios limits the ability to obtain conclusive results, and gender will not be used as a variable.
|