Citation
Effect of Unemployment on Drug Overdose Rates

Material Information

Title:
Effect of Unemployment on Drug Overdose Rates
Creator:
Purcell, Matthew Thomas
Publication Date:
Language:
English

Notes

Abstract:
The CDC estimates from 2000-2015 over half a million Americans have died from drug overdoses. Driving this increase is the ongoing opioid epidemic occurring across the United States. Since 1999, the number of opioid-involved overdoses has quadrupled. States most affected by the opioid epidemic (e.g. West Virginia, Ohio, Kentucky) have also experienced harsh demand shifts on coal and manufacturing labor markets. This has led to summa unemployment in these states. This paper investigates the effects of unemployment on drug overdose rates, using a linear regression model for West Virginia and Florida. West Virginia returned significant results for the unemployment variable, while the Florida model did not. The results of this paper show a connection between unemployment and drug overdose rates in West Virginia. West Virginia represents an epicenter of the opioid epidemic as well as a state that has experience low growth and summa unemployment since the Great Recession. ( en )
General Note:
Awarded Bachelor of Arts, summa cum laude, on May 8, 2018. Major: Economics
General Note:
College or School: College of Liberal Arts and Sciences
General Note:
Advisor: Michelle Phillips. Advisor Department or School: Economics

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Matthew Thomas Purcell. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

UFDC Membership

Aggregations:
UF Undergraduate Honors Theses

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

! Matthew Purcell 7 March 2018 Effect of Unemployment on Drug Overdose Rates I. Introduction The CDC estimates from 2000 2015 over half a million Americans have died from drug overdoses. 1 Drug overdoses in the United States have increased over the last twenty years. Driving this increa se is the ongoing opioid epidemic occurring across the United States. Since 1999 the number of opioid involved overdoses has quadrupled. 2 A higher prescripti on rate has bolstered the increase d use of opioids. From 1999 2014 prescription rates for opioids have increased 356%, showing a clear a ssociation between prescription rates and opioid overdoses. 3 Despite the rise in prescription rates over these years, t here has been no change in the amount of Americans reporting pain. 4 So increased opioid use comes from a change in pain management rather than from an increase in pain presence in society. States most affected by the opioid epidemic (e.g. West Virginia, Ohio, Kentucky) have also experienced harsh demand sh ifts on coal and manufactu ring labor markets. This has le d to high unemployment in these states. 5 These states also report high rates of drug overdoses. This study will look into the effects of unemploym ent rates on drug overdose death rates from West Virginia and Florida counties from 2014 and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

PAGE 2

! A II. Sample The sample for the study comes fr om 63 counties in Florida and 42 in West Virginia from 2014 data and from 63 counties in Florida and 45 countie s in West Virginia from 2015 data The change in the number of West Virginia counties has to do with the availability of prescription opioid rate data. Regressions will be run on Florida data and on West Virginia data separately III. Dependent Variable Drug Overdose Mortality Rate The dependent variable for this study will be the drug overdose death rate. It is the number of drug poisoning deaths per 100,000 population. The data are from the CDC WONDER database. 6 The drug induced deaths included in the data are: 1) nonopioid analgesics, antipyretics and antirheumatics, 2) antiepileptic, sedative hypnotic, antiparkinsonism and psychotropic drugs, not elsewhere classified, 3) narcotics and psychodysleptics [hallucinogens], not elsewhere classified, 4) othe r drugs acting on the autonomic nervous system, and 5) other and unspecified drugs, medicaments and biological substances. Opioids make up 60% of all drug induced deaths, so using the general drug overdose rate can stand as a proxy for the opioid induced d eath rates. 7 IV. Independent Variables Unemployment Rate !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! P #$#:!D/)* E +.(0/(0!;(?/(*!).-.!F;+!*3/)*4/;?;0/5!+*,*.+51!GDHI$2JK:!@-?.(-.%!L@7!#$#%!I.-/;(.?! #*(-*+!F;+!M*.?-1!N-.-/,-/5,O!AB"P:!@<./?.Q?*!.! 1--37889;()*+:5)5:0;< C J=))!J@%!N*-1!S%!$.:);/:;+08"B:"ZZ[Z8449+:44PZZBZ"*"

PAGE 3

! R The unemployment rate is the number of unemployed as a percent of the labor force. The data are from the Bureau of Labor Statistics (BLS). 8 The BLS qualifications for unemployment under the U3 unemployment definition, are that the individual does not have a job, has been on the job search in the last four weeks, and is currently available for work. There has been r esearch into the effects of the opioid cris is on labor force participation, notabl y by Alan Krueger. In his research, he reports people being prescribed opioids were 10% less likely to be employed in 2015 compared to 6% in 1997. 9 This paper suggests that the opioid crisis has contributed to a low labor force participation rate since the Great Recession. 10 Deaton and Case have also shown an increase in "deaths of despair" (i.e. suicide, poisoning) in non college educated whites from 1999 2014. They cite increased unemployment and opioid use as reasons for this decrease in well being. 11 With an increase in the probability of an unemployed individual being prescribed opioids and a decrease in that segments well b eing, it is hypothesized that as unemployment increase s, drug overdose death rates will increase. Rural or Urban Classification A quandary of the opioid epidemic has been its effect on rural vs. urban counties. One study found that for West Virginia and Kentucky citizens in rural areas were more at risk of prescription opioid misuse (POM), while for Florida and New York urban areas !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! [ \]N%!& '(*43?;X4*(-!J.-*,!F;+!N-.-*,%!N*.,; (.??X!@)^=,-*)6 % 1--3,788999:Q?,:0;<8?.=8_5(-X.. % N*3-*4Q*+!AB"C d e+=*0*+%!@?.(: !"#$#%&'(#%)**%+"#%!,$-#$.%/,0#1%)0%20345$6%50+,%+"#%7#8*50#%,9%+"#%:;<;%='>,$%?,$8#% @'$+585A'+5,0%B'+# :! \+;;c/(0,!S.3*+! ;(!25;(;4/5!@5-/
PAGE 4

! T w ere more at risk of POM. 12 One reason for this could be that in states like West Virginia and Kentucky rural unemployment was much higher than urban unemployment, whereas for states like Florida and New York the opposite held true. States with harsher rura l unemployment rates may have been more affected by those labor demand shifts in manufacturing mentioned in the introduction than areas with more urban unemployment. To test this idea, this stud y will use a dummy variable of 1 to denote urban and 0 to den ote rural The US Census definition will be used to differentiate between urban and rural counties. 13 The Census' recorded raw value for a county's population will be used with no adjustment for area of the county. C ounties with a population greater than 50 ,000 will be considered urban, while those l ess than 5 0,000 will be rural. The two regressions are hypothesized to give results here not equal to zero. Since there has been little study on this variable in Florida, the prediction is uncertain. For West Vir ginia, the prediction is a negative value. Median Income Small area income data are found from US Census data. By county, there are data for the median household income level. 14 From 2003 2012 the largest increase in heroin use was with households with an average income between $20,000 and $49,000 (an increase of 77%). 15 Income levels below $20,000 and above $50,000 both saw around a 60% increase in heroin use. Another statistic from the CDC notes that the highest death !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! "A J/00%!e1.+X%!.()!N1.((;(!U;((.-: :$>'0%(.;%B4$'*%7599#$#08#.%50%@$#.8$5A+5,0%CA5,5D%E5.4.#%'F,0G% )D4*+.%50%+"#%:05+#D%<+'+#.H%209,$F50G%B#G5,0%
PAGE 5

! Z rates from prescription opioids wer e in states with the highest rates of poverty. 16 Lower income jobs tend to be more physically demanding and offer less chance for leisure. This could make persons in low income jobs more likely to be prescribed opioids. This study hypothesizes that as incom e decreases, drug overdose death rates will increase. Opioid Prescription Rates As stated in the introduction, the opioid epidemic growth rate mirrors that of the overall opioid prescription rate. So county level prescription rates are taken into accoun t in the study. To do this, CDC data on retail opioid prescriptions dispensed per 100 persons are used. 17 There is high regional variation in prescription rates. The ten percent counties with the highest prescription rates vary by a factor of 31 to one relative to the bottom ten percent. 18 Since 2012 the national prescription rate has been decreasing. 19 This is unsurprising given the rise in awareness of the opioid epidemic in the past few years. The CDC, American Pharmacist Association, American College of Emergency Physicians, and others have updated guidelines for opioid prescribing in response to the growing epidemic. In September of 2017, CVS, one of the largest pharmaceutical chains in the US, announced it would begin to limit opioid prescriptions to seven days and require the use of immediate release formulations before extended opioids. 20 These are all targeted public health measures to control the rates of opioid prescription. An increase in the !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! "P #$#%! 1--3,788999:5)5:0;<8449+83+*+.-* E 4.3,:1-4? @=0=,-! AB"C : "[ e+=*0*+%!@?.(: !"#$#%&'(#%)**%+"#%!,$-#$.%/,0#1%)0%20345$6%50+,%+"#%7#8*50#%,9%+"#%:;<;%='>,$%?,$8#% @'$+585A'+5,0%B'+# : \+;;c/(0,!S.3*+!;(!25;(;4/5!@5-/:1-4? :!AA!N*3-*4Q*+!AB"C:

PAGE 6

! P prescription rate leads to an increase in the supply o f opioids in the community, which in turn fuels either addict ion by the person prescribed the opioid, or misuse by a non prescribed user. The study predicts that an increase in the rate of opioid prescriptions dispens ed will result in an increase in the dr ug overdose death rate. Omitted Variables One omitted variable is a measure of the density of drug rehabilitation centers in a county. With greater a ccess to rehabilitation centers, there is a high chance that users of opioids would turn to professional help before misuse becomes fatal. Florida, as a state, put some of these policies into place (e.g. adding regulated pain clinics) and saw a 50% decrease in the rate of prescription opioid deaths. 21 Therefore, increased density of drug rehabilitation centers would be expected to cause the rate of drug overdose to fall. Along the same lines, a variable that account s for county level policy on and access to overdose prevention drugs, like naloxone, would certainly be helpful for the model. This study could not collect such data, but looking at which counties mandate paramedics to carry such drugs would be a good dummy variable to add to the study. If paramedics were mandated to carry a preventative drug, the study would anticipate a decrease in the overdose deat h rate. Another aspect of the opioid epidemic the model lacks is any sort of measure of the amount of heroin or opioid pills sold on the street. A measure of the black market supply of opioids at the county level would certainly affect the dependent variab le. The intuitive prediction would be that the availability of these types of prescription substitutes would increase misuse and abuse, in turn increasing the drug overdose death rate. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! A" #$#%! 1--3,788999:5)5:0;<8)+=0;<*+);,*83;?/5X8,=55*,,*,:1-4? :!Z!H5-;Q*+!AB" C:

PAGE 7

! C V. Results West Virginia Dataset Summary Statistics (County Level for 2014 and 2015) Variable Observations Mean Standard Deviation Minimum Maximum Drug Overdose Rate (per 100,000 persons ) 87 23.83 14.69 8.00 80.00 Unemployment Rate (U3 ) 87 7.55 2.02 4.10 13.20 Urban/Rural 87 .25 0.44 0 1 Median Income ( Household Level $1) 87 40 666 .00 7 345.54 26 413 .00 67 821 .00 Opioid Prescription Rate (per 100 persons) 87 113.5 0 42.24 27.1 0 263.0 0 Regression 1 R 2 = 0.46 Adjusted R 2 = 0.43 Variable Coefficient Standard Deviation p Value Unemployment Rate 2.98 0.93 0.0020 ** Urban/Rural 5.83 3.51 0.101 Median Income ( Household Level ) 0.0005 0.00023 0.02 Opioid Prescription Rate 0.088 0.034 0.01 The adjusted R 2 for R egression 1 was 0.434. To see if there was interaction between two terms, namely median household income and unemployment, a second regression was run. Income and unemployment were chosen because they seemed to be the most inherently related to each other.

PAGE 8

! [ Regression 2 R 2 = 0.55 Adjusted R 2 = 0.52 Variable Coefficient Standard Deviation p Value Unemployment Rate 1 2. 87 2.58 0.000004 *** Urban/Rural 0.64 3.47 0.85 Median Income ( Household Level ) 0.001 0.000 5 0.01 Opioid Prescription Rate 0. 12 0.03 0.0003 *** Interaction variable (Unemployment and Income) 0.0003 0.00008 0.0001 *** Regression 2 returned an adjusted R 2 of 0.5238, a stronger adjusted R 2 than the original regression. For this reason, it will be looked at for analysis. Unemployment Rate The unemployment rate was found to be significant The null hypothesis that unemployment and drug overdose mortality rate are unrelated can be rejected with 99 .9 % confidence. Also, the interaction term was found to be significant. The null hypothesis that the combined effect of unemployment and income is unrelated to drug overdose mortality rate can be rejected with 99.9% confidence. Taking into account both terms the effect of unemployment would be the coefficient of unemployment (beta unemployment ) plus t he coefficient of the interaction term times the median income (beta interaction *median income). By inputting the mean median income for the data set, the overall effect is that for every 1 percentage point increase in the unemployment rate there is an incr ease of 0.67 drug overdoses per 100,000 persons. By inputting the minimum median income, the

PAGE 9

! d overall effect is that for every 1 percentage point increase in the unemployment rate there is a n increase of 4.95 drug overdoses per 100,000 persons The maximum income yields a result of every 1 percentage point increase in th e unemployment rate there is a decrease of 7.48 drug overdoses per 100,000 persons These results agree with the hypothesis that as unemployment increases the drug overdose mortality rate will increase. Also, with the trend seen by increasing inc ome, there is support that low and mid income jobs experience harsher effects in this crisis. Rural v. Urban This variable was found to be insignificant. Therefo re, the null hypothesis that there is no relationship between classification of a county as rural or urban a nd drug overdose mortality rate cannot be rejected. Median Income The median household income was found to be significant. The null hypothesis th at median household income and drug overdose mortality rate are u nrelated can be rejected with 95 % confidence. As stated before, the interaction term was found to be significant. The null hypothesis that the combined effect of unemployment and income is un related to drug overdose mortality rate can be rejected with 99.9% confidence. Taking into account both terms, the effect of median income would be the coefficient of income (beta income ) plus the coefficient of the interaction term times the unemployment r ate (beta interaction *unemployment). By inputting the mean unemployment rate, the overall effect is that for every dollar increase in the median household income there is a de crease of 0.0013 drug overdoses per 100,000 persons By inputting the minimum unemployment rate, the overall effect is that for every dollar increase in the median household income

PAGE 10

! "B there is a decrease of 0.00023 drug overdoses per 100,000 persons. The maximum unemployment rate yields a result of every dollar increase in the median household income there is a decrease of 0.003 drug overdoses per 100,000 persons. This agrees with the hypothesis that as median household income increase s the drug overdose mortality rate will de crease. Opioid Prescription Rates The o pioid prescription r ate was found to be significant. The null hypothesis that o pioid prescription r ate and drug overdose mortality rate are unrelated can be rej ected with 99.9 % confidence. The coefficient implies that for every 1 percentage point increase in the o pioid prescription r ate there is a n increase of 0. 1204 drug overdoses per 100,000 persons This agrees with the hypothesis that as o pioid prescription r ate increases the drug overdose mortality rate will increase. Florida Dataset Summary Statistics (County Level for 2014 and 2015) Variable Observations Mean Standard Deviation Minimum Maximum Drug Overdose Rate (per 100,000 persons) 126 15.21 5.87 4.00 31.00 Unemployment Rate (U3) 126 6.18 1.19 3.50 11.30 Urban/Rural 126 .65 0.48 0 1 Median Income ( Household Level $1) 126 45 788 .00 8 349.8 2 32 351 .00 71 896 .00 Opioid Prescription Rate (per 100 persons) 126 83.69 28.0 2 4.90 184.10 Regression 3 R 2 = 0.09 Adjusted R 2 = 0.06

PAGE 11

! "" Variable Coefficient Standard Deviation p Value Unemployment Rate 0.2 7 0.50 0.60 Urban/Rural 2.3 1 1.40 0.10 Median Income ( Household Level ) 0.0 00007 0.0 0009 0.9 4 Opioid Prescription Rate 0.05 0.02 0.0085 ** The adjusted R 2 for this regression was 0.06106. To see, again, if there was interaction between two terms, namely median household income and unemployment, a second regression was run. Income and unemployment were chosen for similar reason as for the West Virginia regre ssion. Regression 4 R 2 = 0.10 Adjusted R 2 = 0.06 Variable Coefficient Standard Deviation p Value Unemployment Rate 2.86 2.33 0.22 Urban/Rural 2.11 1.41 0.14 Median Income ( Household Level ) 0.0003 0.0003 0.27 Opioid Prescription Rate 0. 05 0.02 0.01 Interaction variable (Unemployment and Income) 0.00006 0.00005 0.26 There was no improvement seen in the R 2 of the interaction model. So R egression 3 will be discussed. Unemployment Rate

PAGE 12

! "A This variable was found to be insignificant. Therefore, the null hypothesis that there is no relationship between unemployment and drug overdose mortality rate cannot be rejected. Rural v. Urban This variable was found to be insignificant. Therefore, the null hypothesis that there is no relationship between classification of a county as rural or urban and drug overdose mortality rate cannot be rejected. Median Income This variable was found to be insignificant. Therefore, the null hypothesis that there is no relationship between median household income and drug overdose mortality rate cannot be rejected. Opioid Prescription Rates The o pioid prescription r ate was found to be significant. The null hypothesis that o pioid prescription r ate and drug overdos e mortality rate are unr elated can be rejected with 99 % confidence. The coefficient implies that for every 1 percentage point increase in the o pioid prescription r ate there is a 0.0488 percentage point increase in the drug overdose mortality rate. This agr ees with the hypothesis that as o pioid prescription r ate increases the drug overdose mortality rate will increase. VI. Conclusion The results of this paper show a connection between unemployment and drug overdose rates in West Virginia. West Virginia repre sents an epicenter of the opioid epidemic as well as a state that has experience low growth and high unemployment since the Great Recession. These facts reflect in the R 2 of 0.5238 Regression 2 returned.

PAGE 13

! "R Florida represents a different environment. While the opioid epidemic certainly is an issue for the state, Florida has had a much stronger economic re covery from the Great Recession. This caused drug overdose rates to have different root causes than the ones this paper's models control for. This is presumably why Florida's R 2 was 0.06106. The results of this paper show a clear connection be tween opioid prescription rates. Perhaps, this is good news because of all the varia bles in this paper opioid prescription rates seem to be the easiest one for policy measures to tackle. As discussed earlier in the paper, many states have already begun efforts to manage prescription rates. Florida is one, and its policy changes have led t o a decrease in oxycodone overdose deaths of over 50%. 22 More research into which state policy have had the best results in controlling prescription rates would inform other states and federal policymakers on what measures to take to best abate the epidemic Finally, drug overdoses are a public health issue with its roots in addiction. Unlike a pathogenic epidemic, drug related epidemics manifest and spread by humans behavior alone. Human behavior created this issue, and with enough study and research behav iors can change, bringing an end to a crisis. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! AA #$#%! 1--3,788999:5)5:0;<8)+=0;<*+);,*83;?/5X8,=55*,,*,:1-4? :!Z!H5-;Q*+!AB"C:!