近二十年間，有許多研究在探討抽菸與就業之間的關係，其中多數的研究發現抽菸之盛行率在失業人口中較高；有一些文獻則指出失業可能會導致較高的抽菸機率以及增加抽菸的香菸數；只有少數的文章在反向研究抽菸或是戒菸對於就業之影響，特別是，目前尚未有此類研究建立抽菸或是戒菸對於就業的因果關係。

    此篇研究使用三元probit模型分析美國2017 Behavioral Risk Factor Surveillance System的資料，並利用抽菸相關行為的州盛行率、個人安全帶使用狀況以及個人最大攝氧量作為工具變數，以處理潛在之內生性問題並建立抽菸、戒菸對於就業之因果關係。

    此研究之貢獻在於，三元probit模型之分析結果顯示抽菸與就業之間的確存在內生性，而在考量內生性後，對女性樣本而言，抽菸會導致就業的機率下降0.14個百分點，對男性而言則是下降0.07百分點。此外分析結果指出，對女性樣本而言，戒菸會使得就業的機率上升0.05個百分點，就我們目前的了解而言，此效果目前尚未有其他文獻紀載或研究，就男性而言，戒菸對於就業沒有顯著影響。從此篇研究的結果而言，若從成本效益分析來切入戒菸政策，綜合考量男女抽菸行為對於就業的負向影響後，女性之戒菸的正向影響也不容忽視。

Over these two decades, there have been many literature discussing the relationship between smoking and employment. Most found a higher prevalence of smoking among the unemployed. Some found being unemployed could increase the intensity of smoking and the probability of being a smoker. Few study the association between being a current or former smoker and employment. In particular, there hasn’t been a causal study regarding how being a current or former smoker could affect employment.

This paper uses the trivariate probit model with instrumental variables, the prevalence of smoking by each state, seatbelt use and the maximal oxygen uptake, to estimate the causal effect of smoking status on employment and to take the potential endogeneity into account. After examining the data from 2017 Behavioral Risk Factor Surveillance System, an annual health-related telephone surveys conducted in all 50 states in U.S, we found being a current smoker would lead to a 0.14 percentage point deduction of the probability of being employed for females and 0.07 for males compared to those who never smoke. Another contribution of the paper is that we found that for a female observation to be a former smoker would increase the probability of being employed by 0.05 percentage points compared to those who never smoke while for male there’s no significant impact. To the best of our knowledge, the causal effects mentioned above have not yet been documented before. Also, our results suggest there do exist endogeneity between smoking status and employment. Therefore, the using of trivariate probit model with instrumental variables is desirable.

Following our analysis, considering being a current smoker’s negative impact on employment for both females and males, the benefit on employment of females being a former smoker is important from the cost-benefit perspective of the smoking cessation policy.

Table of Contents
I.  Introduction	1
II. Data	        3
III.Empirical Model	5
IV. Results	        6
V.  Conclusion:	        20
Reference	        21
Appendix A	        23
Appendix B	        28



List of Tables
TABLE 1 SMOKING AND EMPLOYMENT STATUS BY GENDER	4
TABLE 2 EMPLOYMENT RATE BY SMOKING STATUS AND GENDER	5
TABLE 3 UNIVARIATE PROBIT MODEL ON EMPLOYMENT	6
TABLE 4 TTRIVARIATE PROBIT MODEL ON EMPLOYMENT FOR FEMALES: USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	8
TABLE 5 TTRIVARIATE PROBIT MODEL ON EMPLOYMENT FOR MALES: USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	10
TABLE 6 MARGINAL EFFECTS OF SMOKING STATUS ON EMPLOYMENT. RESULTS BASED ON TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	12
TABLE 7 PAIRWISE CORRELATION BETWEEN THE RESIDUALS FROM THE THREE EQUATIONS OF THE TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	12
TABLE 8 TEST OF INSTRUMENTAL VARIABLE RELEVANCE AND EXCLUSION FOR TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	13
TABLE 9 TTRIVARIATE PROBIT MODEL ON EMPLOYMENT FOR FEMALES: USING THE MAXIMAL OXYGEN UPTAKE AND DOES NOT ALWAYS OR DOES NOT NEARLY ALWAYS USING SEATBELT AS INSTRUMENTAL VARIABLES	14
TABLE 10 TTRIVARIATE PROBIT MODEL ON EMPLOYMENT FOR MALES: USING THE MAXIMAL OXYGEN UPTAKE AND DOES NOT ALWAYS OR DOES NOT NEARLY ALWAYS USING SEATBELT AS INSTRUMENTAL VARIABLES	16
TABLE 11 MARGINAL EFFECTS OF SMOKING STATUS ON EMPLOYMENT. RESULT BASED ON TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	17
TABLE 12 PAIRWISE CORRELATION BETWEEN THE RESIDUALS FROM THE THREE EQUATIONS OF THE TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	18
TABLE 13 TEST OF INSTRUMENTAL VARIABLE RELEVANCE AND EXCLUSION FOR TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	19

	                                       


Appendix A

TABLE A 1 TTRIVARIATE PROBIT MODEL OF SMOKING STATUS ON EMPLOYMENT FOR FEMALES: USING THE GENDER SPECIFIC PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	24
TABLE A 2 TTRIVARIATE PROBIT MODEL OF SMOKING STATUS ON EMPLOYMENT FOR MALES: USING THE GENDER SPECIFIC PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	26
TABLE A 3 MARGINAL EFFECTS OF SMOKING STATUS ON EMPLOYMENT. RESULTS BASED ON TRIVARIATE PROBIT ESTIMATION USING THE GENDER SPECIFIC PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	27
TABLE A 4 PAIRWISE CORRELATION BETWEEN THE RESIDUALS FROM THE THREE EQUATIONS OF THE TRIVARIATE PROBIT ESTIMATION USING THE GENDER SPECIFIC PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	28
TABLE A 5 TEST OF INSTRUMENTAL VARIABLE RELEVANCE AND EXCLUSION FOR TRIVARIATE PROBIT ESTIMATION USING THE GENDER SPECIFIC PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	28


Appendix B

TABLE B 1 MARGINAL EFFECTS OF SMOKING STATUS ON EMPLOYMENT. RESULT BASED ON TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES.	29
TABLE B 2 PAIRWISE CORRELATION BETWEEN THE RESIDUALS FROM THE THREE EQUATIONS OF THE TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	29
TABLE B 3 TEST OF INSTRUMENTAL VARIABLE RELEVANCE AND EXCLUSION FOR TRIVARIATE PROBIT ESTIMATION USING THE PREVALENCE OF NEVER SMOKER AND FORMER SMOKER AS INSTRUMENTAL VARIABLES	30

1.	Arcaya, M., Glymour, M. M., Christakis, N. A., Kawachi, I., & Subramanian, S. V. (2014). Individual and spousal unemployment as predictors of smoking and drinking behavior. Social Science & Medicine, 110, 89-95.
2.	Becker, N. (2003). Epidemiological aspects of cancer screening in Germany. Journal of cancer research and clinical oncology, 129(12), 691-702.
3.	Cappellari, L., Jenkins, S.P. 2003. "MVPROBIT: Stata module to calculate multivariate probit regression using simulated maximum likelihood," Statistical Software Components S432601, Boston College Department of Economics, revised 25 Jan 2006. https://ideas.repec.org/c/boc/bocode/s432601.html
4.	Danaei, G., Ding, E. L., Mozaffarian, D., Taylor, B., Rehm, J., Murray, C. J., & Ezzati, M. (2009). The preventable causes of death in the United States: comparative risk assessment of dietary, lifestyle, and metabolic risk factors. PLoS medicine, 6(4), e1000058. https://doi.org/10.1371/journal.pmed.1000058. 
5.	De Vogli, R., & Santinello, M. (2005). Unemployment and smoking: does psychosocial stress matter?. Tobacco control, 14(6), 389-395.
6.	Jamal, A., Phillips, E., Gentzke, A. S., Homa, D. M., Babb, S. D., King, B. A., & Neff, L. J. (2018). Current cigarette smoking among adults—United States, 2016. Morbidity and Mortality Weekly Report, 67(2), 53.
7.	Kendzor, D. E., Reitzel, L. R., Mazas, C. A., Cofta-Woerpel, L. M., Cao, Y., Ji, L., ... & Castro, Y. (2012). Individual-and area-level unemployment influence smoking cessation among African Americans participating in a randomized clinical trial. Social science & medicine, 74(9), 1394-1401.
8.	Kriegbaum, M., Larsen, A. M., Christensen, U., Lund, R., & Osler, M. (2011). Reduced probability of smoking cessation in men with increasing number of job losses and partnership breakdowns. Journal of Epidemiology & Community Health, 65(6), 511-516.
9.	Merline, A. C., O’Malley, P. M., Schulenberg, J. E., Bachman, J. G., & Johnston, L. D. (2004). Substance use among adults 35 years of age: prevalence, adulthood predictors, and impact of adolescent substance use. American Journal of Public Health, 94(1), 96-102
10.	Prochaska, J. J., Shi, Y., & Rogers, A. (2013). Tobacco use among the job-seeking unemployed in California. Preventive medicine, 56(5), 329-332.
11.	Unemployment Rates for States, 2017 Annual Averages. (n.d.). Retrieved July 16, 2019, from https://www.bls.gov/lau/lastrk17.htm
12.	World Health Organization (n.d.) Retrieved June 20, 2019, from http://apps.who.int/gho/data/node.sdg.3-a-viz?lang=en