Are Graphic Warning Labels Stopping Millions of Smokers? a Comment on Huang, Chaloupka, and Fong
Total Page:16
File Type:pdf, Size:1020Kb
APPENDICES FOR ARE GRAPHIC WARNING LABELS STOPPING MILLIONS OF SMOKERS? A COMMENT ON HUANG, CHALOUPKA, AND FONG (ECON JOURNAL WATCH, MAY 2018) Trinidad Beleche* Nellie Lew* Rosemarie L. Summers* J. Laron Kirby* * United States Food and Drug Administration (FDA), 10903 New Hampshire Ave, Silver Spring, MD 20993, United States. Corresponding author: Trinidad Beleche, E-mail: [email protected]. This article reflects the views of the authors and should not be construed to represent FDA’s views or policies. We thank Don Kenkel, Clark Nardinelli, and Ian Irvine for their comments on an earlier version of the manuscript. All errors are our own. 1 Appendix Figure A1. — Residuals and Fitted Values under Various Specifications Model 4: Canada, Logged Trends Model 4: Canada, Non-logged Trends Model 4: US, Logged Trends Model 4: US, Non-logged Trends .1 .04 .05 .04 .02 .05 .02 0 0 0 0 -.02 -.05 -.05 -.02 -.04 -.1 -.04 -.1 -.06 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 Year Year Year Year Residuals Fitted values Residuals Fitted values Residuals Fitted values Residuals Fitted values Panel A. —Model 4: Residuals and Fitted Panel B. —Model 4: Residuals and Fitted Values, Canada Values, U.S. Model 5: Canada, Logged Trends Model 5: Canada, Non-logged Trends Model 5: US, Logged Trends Model 5: US, Non-logged Trends .04 .05 .04 .05 .02 .02 0 0 0 0 -.02 -.05 -.05 -.02 -.04 -.1 -.1 -.04 -.06 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 Year Year Year Year Residuals Fitted values Residuals Fitted values Residuals Fitted values Residuals Fitted values Panel C. —Model 5: Residuals and Fitted Panel D.--Model 5: Residuals and Fitted Values, Canada Values, U.S. Model 6: Canada, Logged Trends Model 6: Canada, Non-logged Trends Model 6: US, Logged Trends Model 6: US, Non-logged Trends .1 .04 .04 .04 .02 .02 .05 .02 0 0 0 0 -.02 -.02 -.04 -.05 -.02 -.04 -.06 -.06 -.1 -.04 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 Year Year Year Year Residuals Fitted values Residuals Fitted values Residuals Fitted values Residuals Fitted values Panel E. —Model 6: Residuals and Fitted Panel F. —Model 6: Residuals and Fitted Values, Canada Values, U.S. 2 Appendix Figure A2. —Trends of Normalized Excise Tax, Normalized Price Indices for Canada and the U.S. Panel A. — Trends of Normalized Excise Tax Panel B. —Trends of Normalized Price Index for Canada and the U.S. with ITC Price for Canada and the U.S. 8 8 6 6 4 4 Excise Tax (Normalized)Excise Tax 2 2 Price(Normalized)ITC with 0 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Canada US Canada US Panel C. —Trends of Normalized Index Price without ITC Price for Canada and the U.S. 8 6 4 2 Price without (Normalized)ITC 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Canada US 3 Appendix Figure A3. — Log of Smoking Rates in Canada and the United States, 1991–2009 3.4 3.2 3 Smoking(Log) Rate 2.8 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Canada US 4 Appendix Table A1. —Graphic Warning Labels on Cigarette Packs by Country United Arab Emirates Canada (2001)* Pakistan (2010) Bulgaria (2016) (2012) Czech Republic Brazil (2002)* Macedonia (2010) Hungary (2012)* (2016) Switzerland Singapore (2004)* Macau (2013) Estonia (2016) (2010)** Liechtenstein Thailand (2005)* Iceland (2013) Finland (2016) (2010)** Venezuela (2005)* Mongolia (2010)* Ireland (2013)* Germany (2016) Jordan (2006)* Colombia (2010)* Russia (2013)* Greece (2016) Australia (2006)*,** Turkey (2010) Kazakhstan (2013)* Italy (2016) Uruguay (2006)* Mexico (2010)* Seychelles (2013)* Lithuania (2016) Panama (2006)* Norway (2011) Fiji (2013) Netherlands (2016) Belgium (2006)*,** Malta (2011)* Vietnam (2013) Poland (2016) Chile (2006)* France (2011)* Montenegro (2013) Portugal (2016) Hong Kong (2007)* Guernsey (2011) Indonesia (2014) San Marino (2016) New Zealand Spain (2011) Nepal (2014)* Slovakia (2016) (2008)*,** Romania (2008) Bolivia (2011)* Costa Rica (2014)* Sweden (2016) United Kingdom Jersey (2012) Jamaica (2014) Cambodia (2016) (2008)* Egypt (2008)* Ukraine (2012) Suriname (2014) Bangladesh (2016) Brunei (2008)* Honduras (2012)* Yemen (2014) South Korea (2016) Madagascar Cook Islands (2008) Samoa (2014) Kenya (2016) (2012)* Iran (2009) Denmark (2012)* Sri Lanka (2015) Greenland (2016) Malaysia (2009)* Ecuador (2012)* Solomon Islands (2015) Khartoum (2016) Taiwan (2009)* Argentina (2012)* Turkmenistan (2015) Kyrgyzstan (2016) El Salvador Peru (2009)* Namibia (2015) Vanautu (2017) (2012)* Djibouti (2009) Bahrain (2012) Chad (2015) Armenia (2017) Mauritius (2009) Kuwait (2012) Philippines (2016) Belarus (2017) India (2009)* Oman (2012) Lao (2016) Cayman Islands Qatar (2012) Myanmar (2016) (2009) Saudi Arabia Latvia (2010)* Austria (2016)* (2012) Notes: * Countries have implemented other rounds of graphic warnings. ** Countries rotate sets every 12 months. Source: Canadian Cancer Society (2016) 5 Appendix Table A2.—Adding Year Fixed Effects to the Specifications Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Panel A. Huang et al.’s Specification: Logged Monthly Trends, Non-robust Standard Errors CA*GWL −0.186*** −0.182*** −0.186*** −0.065 −0.058 −0.062 (0.046) (0.044) (0.044) (0.041) (0.041) (0.041) Panel B. Huang et al.’s Specification: Logged Monthly Trends, Robust Standard Errors CA*GWL −0.186*** −0.182*** −0.186*** −0.065*** −0.058** −0.062*** (0.052) (0.048) (0.048) (0.019) (0.019) (0.019) Panel C. Huang et al.’s Specification: Non-logged Monthly Trends, Robust Standard Errors CA*GWL −0.192*** −0.191*** −0.195*** −0.087*** −0.075*** −0.081*** (0.051) (0.046) (0.046) (0.019) (0.014) (0.013) Controls: Ln/Monthly Trend Yes Yes Yes Yes Yes Yes Ln(Index Excise Tax) Yes Yes Ln(Price Index w/o Yes Yes ITC Price) Ln(Price Index w ITC Yes Yes Price) Canada & Trend Yes Yes Yes Interaction Year Fixed Effects Yes Yes Yes Yes Yes Yes Notes: Unless specified, robust standard errors are reported in parentheses. *p<0.10, **p<0.05, ***p<0.01. The list of Controls applies to panels A to C. Panels A and B control for logged monthly trends. Panel C controls for non-logged monthly trends. All specifications control for year fixed effects. 6 Appendix Table A3.—Robustness Checks Using Logged Trends Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Panel A. Falsification Test of Huang et al.’s Specification where Outcome is Fatal Car Accidents CA*GWL −0.052 −0.067 −0.082* −0.051 −0.087* −0.114** (0.051) (0.046) (0.040) (0.067) (0.048) (0.043) Panel B. Outcome is Smoking rate, Treatment is 1999–2000, Control is 1995–1998 CA*GWL −0.102** −0.076 −0.076 −0.069* −0.094** −0.094** (0.025) (0.044) (0.044) (0.024) (0.028) (0.028) Panel C. Outcome is Smoking rate, Treatment is 1996–2000, Control is 1991–1995 CA*GWL −0.107** −0.135** −0.135** 0.141 0.216* 0.216* (0.034) (0.041) (0.041) (0.072) (0.098) (0.098) Controls: Ln(Monthly Trend) Yes Yes Yes Yes Yes Yes Ln(Index Excise Tax) Yes Yes Ln(Price Index w/o Yes Yes ITC Price) Ln(Price Index w ITC Yes Yes Price) Canada & Trend Yes Yes Yes Interaction Notes: Robust standard errors in parentheses. *p<0.10, **p<0.05, ***p<0.01. Outcome is log of annual number of fatal car accidents. Sources: Canadian Council of Motor Transport Administrators (2011); National Highway Traffic Safety Administration (1995; 2017). 7 Appendix B. Allowing For Changes in Policy Effects Over Time: A Dynamic Specification In econometric modeling, it is important to control for confounding factors or variables that are correlated with both the dependent and the independent variable. Failure to control for such factors can eliminate or cast doubt on the internal validity of the estimates. In the case of the difference-in-difference approach, it is important to consider whether a policy has a differential effect on an outcome over time. If so, country-specific time trends may make it difficult to disentangle the effects of pre-existing trends and the effects of the policy intervention. Huang et al. point out that their “estimated impact of GWLs on smoking rates is the average impact over the 2001 to 2009 period. The impact of GWLs may erode over time as smokers become inured to the labels and the novelty of the GWLs wear[s] off. Future studies could improve our analyses by accounting for other tobacco control policies and other factors that could influence smoking rates in Canada and the USA” (2014, i11). As argued by Justin Wolfers, “any reduced-form or structural analysis that assumes an immediate constant response to a policy shock may be mis-specified if actual dynamics are more complex than a simple series break” (2006, p. 1807). That is, a single variable indicating implementation of graphic warning labels (namely, GWL*PostGWL as shown in equation (1) in the article) would not capture the effects of the policy over time (i.e., short-run vs long-run). Therefore, we estimate a modified version of equation (1) to allow the GWL impact to vary over time, and estimate equation (B1) using the non-logged trend instead: 8 (B1) ln(푆푚표푘푛푔푅푎푡푒) = 훼 + 훽1퐶푎푛푎푑푎 + 훽2퐶푎푛푎푑푎 ∗ 푃푟푒퐺푊퐿푂푛푒푡표퐹표푢푟 + 훽3퐶푎푛푎푑푎 ∗ 푃푟푒퐺푊퐿퐹푣푒푡표퐸푔ℎ푡 + 훽4퐶푎푛푎푑푎 ∗ 푌푒푎푟표푓퐺푊퐿 + 훽5퐶푎푛푎푑푎 ∗ 푃표푠푡퐺푊퐿푂푛푒푡표퐹표푢푟 + 훽6퐶푎푛푎푑푎 ∗ 푃표푠푡퐺푊퐿퐹푣푒푡표퐸푔ℎ푡 + 훽7(푙푛퐸푥푐푠푒 푇푎푥 표푟 푃푟푐푒 퐼푛푑푒푥) + 훽8푀표푛푡ℎ푙푦푇푟푒푛푑 + 훽9퐶푎푛푎푑푎 ∗ 푀표푛푡ℎ푙푦푇푟푒푛푑 + 푒푟푟표푟 Equation (B1) includes lead and lag variables for the policy (henceforth we refer to it as the dynamic model) to identify the policy response function.