Price elasticity of demand of non- products: a systematic review and meta-analysis

Mohammed Jawad, John Tayu Lee, Stanton Glantz, Christopher Millett

APPENDICES

Appendix 1. Electronic search strategy (conducted on 18th November 2017)

Search terms (in all fields) Number retrieved Medline Embase EconLit Web of (1946 (1947 (1971 Science to to to (1970 to present) present) present) present) 1. (commerce OR tax OR cost OR price OR “price 541,683 823,555 389,150 1,506,788 elastic*”) 2. (arghile* OR argile* OR * OR “betel quid” OR 26,505 16,502 565 10,862 bidi* OR calean OR OR OR cigarillo* OR cheroot* OR “chewing tobacco” OR chillum* OR chutta* OR dhumti* OR “dipping tobacco” OR “dissolvable tobacco” OR dokha OR “double corona” OR “e-cig*” OR “e cig*” OR “electronic cig*” OR goza OR gundi* OR gutka OR gutkha OR hogesoppu OR hookah OR hookli OR “hubble bubble” OR “hubbly bubbly” OR huqqa* OR kadapam* OR kaddipudi* OR kalyan OR khaini OR kiwam OR kizami OR * OR mawa OR medwakh OR midwakh OR mishri OR narghile* OR nargile* OR nass OR naswa OR paan OR “paan masala” OR “pan masala” OR pattiwala OR “pipe tobacco” OR qalyan OR qiwam OR “roll-your-own” OR RYO OR shammah OR sheesha OR shisha OR snuff OR snus OR stumpen OR suipa OR “tobacco gum” OR “tobacco water” OR waterpipe OR water-pipe OR “water pipe” OR zarda) 3. (1 AND 2) 603 816 173 434

Appendix 2. Data abstraction

Study and Study features Statistical approach Elasticity of demand year (95% confidence interval) Bidis Guindon  Country: India  Model type: two-equation system  Own price (bidi): 2011  Time period: 1993 to of budget shares and unit values -0.940 (-0.979, -0.901) 2005 to correct for quality and  Design: repeat cross- measurement error  Cross-price (bidi vs. sectional and pooled  Unit of consumption: quantity ): cross-sectional consumed in the last 30 days -0.021 (-0.048, 0.006)  Price type: nominal  Model adjustments:  Demand data source: o Income: yes National Sample o Cigarette price/tax: yes Survey Organisation o Tobacco policy: no (NSS) o Sociodemographics: yes  Price/tax data source: National Sample Survey Organisation (NSS)Population: adults John 2008  Country: India  Model type: two-equation system  Own price (bidi,  Time period: 1999 to of budget shares and unit values urban): 2000  Unit of consumption: quantity -0.855 (-1.020, -0.690)  Design: cross- consumed in the last 30 days sectional  Model adjustments:  Own price (bidi,  Demand data source: o Income: yes rural): National Sample o Cigarette price/tax: yes -0.922 (-1.006, -0.838) Survey Organisation o Tobacco policy: no (NSS) (N=120,309) o Sociodemographics: yes  Cross-price (bidi vs.  Price type: nominal cigarettes, urban):  Price/tax data source: -0.082 (-0.136, -0.028) National Sample Survey Organisation  Cross-price (bidi vs. (NSS) (N=120,309) cigarettes, rural):  Population: adults -0.063 (-0.245, 0.119) Joseph 2014  Country: India  Model type: probit model  Own price (bidi):  Time period: 1999 to  Unit of consumption: current bidi -2.693*** 2004 use  Design: cross-  Model adjustments: sectional o Income: yes  Demand data source: o Cigarette price/tax: yes Global Youth o Tobacco policy: no Tobacco Survey o Sociodemographics: yes  Price type: nominal  Price/tax data source: Global Youth Tobacco Survey  Population: 13 to 15 years old Nargis 2010  Country: Bangladesh  Model type: logit model, log-  Own price (bidi,  Time period: 2009 linear model participation):  Design: cross-  Unit of consumption: -0.460 sectional participation, intensity  Demand data source:  Model adjustments:  Own price (bidi, ITC Bangladesh o Income: no intensity): Survey (N=83,356) o Cigarette price/tax: yes -0.180  Price type: not o Tobacco policy: no reported o Sociodemographics: yes  Price/tax data source: census and ITC Bangladesh Survey  Population: adults Selvaraj  Country: India  Model type: two-equation system  Own price (bidi, low 2015  Time period: 2011 to of budget shares and unit values income): 2012  Unit of consumption: quantity -0.433 (-0.437, -0.428)  Design: cross- consumed in the last 30 days sectional  Model adjustments:  Own price (bidi,  Demand data source: o Income: yes middle income): Consumer o Cigarette price/tax: yes -0.250 (-0.254, -0.246) Expenditure Survey o Tobacco policy: no (CES) o Sociodemographics: yes  Own price (bidi, high  Price type: nominal income):  Price/tax data source: -0.082 (-0.087, -0.076) Consumer Expenditure Survey  Cross price (bidi vs. (CES) cigarettes, low  Population: adults income): 0.013 (0.007, 0.019)

 Cross price (bidi vs. cigarettes, middle income): -0.039 (-0.045, -0.033)

 Cross price (bidi vs. cigarettes, high income): -0.111 (-0.117, -0.105)

Shang 2017  Country: India  Model type: Generalised  Own price (bidi):  Time period: 2010 to estimating equation -0.013 (-0.062, 0.036) 2013  Unit of consumption: current use  Design: repeat cross-  Model adjustments: sectional o Income: yes  Demand data source: o Cigarette price/tax: no India TCP Survey o Tobacco policy: yes  Price type: tax o Sociodemographcis: yes  Price/tax data source: Campaign for Tobacco Free Kids and ERC report of tobacco market Cigars Ciccarelli  Country: Italy  Model type: dynamic time series  Own price (cigars): 2014  Time period: 1871 to model -0.293 (-0.369, -0.217) 1913  Unit of consumption:  Design: time series consumption of cigars  Demand data source:  Model adjustments: annual budget reports o Income: yes of o Cigarette price/tax: yes management o Tobacco policy: no companies o Sociodemographics: yes  Price type: real  Price/tax data source: annual budget reports of tobacco industry management companies  Population: consumers Da Pra 2009  Country: US  Model type: log-linear model  Own price (cigars):  Time period: 2006 to  Unit of consumption: sales of 0.980 2008 cigars  Design: time series  Model adjustments:  Cross-price (cigars vs.  Demand data source: o Income: no cigarettes): store scanner data o Cigarette price/tax: yes 0.120  Price type: nominal o Tobacco policy: no  Price/tax data source: o Sociodemographics: no store scanner dataPopulation: consumers Escario  Country: Spain  Model type: dynamic time series  Own price (cigars): 2004  Time period: 1964 to model -0.932 (-1.482, -0.382) 1995  Unit of consumption:  Design: time series consumption of cigars  Cross-price (cigars vs.  Demand data source:  Model adjustments: cigarettes [black Spanish State o Income: no tobacco]): Tobacco Company, o Cigarette price/tax: yes 1.111 (0.627, 1.595) National Accounts o Tobacco policy: yes (OECD) o Sociodemographics: no  Cross-price (cigars vs.  Price type: nominal cigarettes [Virginia  Price/tax data source: tobacco]): Spanish State -1.221 (-1.848, -0.595) Tobacco Company, National Accounts (OECD)  Population: not reported Lee 2005  Country: Taiwan  Model type: dynamic time series  Own price (cigars):  Time period: 1971 to model -0.047 (-0.815, 0.721) 2000  Unit of consumption: number of  Design: time series cigars consumed  Demand data source:  Model adjustments: Taiwan Tobacco and o Income: yes Wine Monopoly o Cigarette price/tax: yes Bureau (TTWMB) o Tobacco policy: no  Price type: real o Sociodemographics: no  Price/tax data source: average retail price per piece weighted by the market sale quantity of each  Population: 15 years+ Pekurinen  Country: Finland  Model type: dynamic time series  Own price (cigars): 1989  Time period: 1960 to model -1.697 (-2.205, -1.189) 1987  Unit of consumption: number of  Design: time series cigars consumed  Demand data source:  Model adjustments: Central Statistical o Income: yes Office o Cigarette price/tax: no  Price type: real o Tobacco policy: yes  Price/tax data source: o Sociodemographics: no Central Statistical Office  Population: 15 years+ Ringel 2005  Country: US  Model type: logistic regression  Own price (cigars):  Time period: 1999 to  Unit of consumption: current use -0.336 2000 (participation)  Design: repeat cross-  Model adjustments: sectional o Income: no  Demand data source: o Cigarette price/tax: yes National Youth o Tobacco policy: no Tobacco Survey o Sociodemographics: yes  Price type: nominal  Price/tax data source: market level grocery scanner price information on cigars (dollar per cigar), obtained from marketing firm ACNielsen  Population: high school students Zheng 2016  Country: US  Model type: ordinary least  Own price (cigars):  Time period: 2009 to squares (OLS) regression -1.108 (-1.210, -1.006) 2013  Unit of consumption: sales of  Design: time series cigars  Cross-price (cigars vs.  Demand data source:  Model adjustments: cigarettes): retail scanner data o Income: yes -0.001 (-0.007, 0.005)  Price type: nominal o Cigarette price/tax: yes  Price/tax data source: o Tobacco policy: no retail scanner data o Sociodemographics: yes  Population: consumers Zheng 2017  Country: US  Model type: Almost Ideal  Own price (cigars):  Time period: 2009 to Demand System -1.501 2013  Unit of consumption: sales of  Design: time series cigars  Cross-price (cigars  Demand data source:  Model adjustments: vs. cigarettes): retail scanner data o Income: yes -0.210  Price type: nominal o Cigarette price/tax: yes  Price/tax data source: o Tobacco policy: no retail scanner data o Sociodemographics: yes  Population: consumers Little cigars Gammon  Country: US  Model type: log-log demand  Own price (little 2015  Time period: 2011 to model cigars): 2013  Unit of consumption: sales of -3.170 (-3.836, -2.504)  Design: time series little cigars  Demand data source:  Model adjustments:  Cross-price (little retail scanner data o Income: yes cigars vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes 2.730 (1.103, 4.357)  Price/tax data source: o Tobacco policy: yes retail scanner data o Sociodemographics: yes  Population: consumers Zheng 2016  Country: US  Model type: ordinary least  Own price (little  Time period: 2009 to squares (OLS) regression cigars): 2013  Unit of consumption: sales of -0.886 (-1.074, -0.698)  Design: time series little cigars  Demand data source:  Model adjustments:  Cross-price (little retail scanner data o Income: yes cigars vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes 0.019 (0.007, 0.031)  Price/tax data source: o Tobacco policy: no retail scanner data o Sociodemographics: yes  Population: consumers Zheng 2017  Country: US  Model type: Almost Ideal  Own price (little  Time period: 2009 to Demand System cigars): 2013  Unit of consumption: sales of -1.428  Design: time series little cigars  Demand data source:  Model adjustments:  Cross-price (little retail scanner data o Income: yes cigars vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes -0.170  Price/tax data source: o Tobacco policy: no retail scanner data  Sociodemographics: yes  Population: consumers Electronic cigarettes Huang 2014  Country: US  Model type: ordinary least  Own price (e-  Time period: 2009 to squares (OLS) regression cigarettes, disposable): 2012  Unit of consumption: sales of e- -2.135 (-3.768, -0.502)  Design: time series cigarettes  Demand data source:  Model adjustments:  Own-price (e- commercial retail o Income: yes cigarettes, reusable): store scanner data o Cigarette price/tax: yes -1.829 (-2.791, -0.867)  Price type: real o Tobacco policy: yes  Price/tax data source: o Sociodemographics: yes  Cross-price (e- commercial retail cigarette, disposable store scanner data vs. cigarettes):  Population: 0.535 (-0.055, 1.125) consumers  Cross-price (e- cigarette, reusable vs. cigarettes): 1.217 (-0.394, 2.828)

Pesko 2017  Country: US  Model type: GLM and logit  Own price (e-  Time period: 2014 to model cigarettes, disposable, 2015  Unit of consumption: participation):  Design: repeat cross- participation and intensity of e- -0.649 (-1.431, 0.133) sectional cigarette use  Demand data source:  Model adjustments:  Own price (e- Monitoring the Future o Income: yes cigarettes, disposable, surveys o Cigarette price/tax: yes intensity):  Price type: nominal o Tobacco policy: yes -0.974 (-1.769, -0.178)  Price/tax data source: o Sociodemographics: yes commercial retail  Own price (e- store scanner data cigarettes, refills,  Population: youth participation): -0.426 (-1.759, 0.908)

 Own price (e- cigarettes, refills, intensity): 0.711 (-0.514, 1.936)

Stoklosa  Country: Estonia,  Model type: ordinary least  Own price (e- 2016 Ireland, Latvia, squares (OLS) regression cigarettes): Lithuania, Sweden,  Unit of consumption: per capita -0.780 United Kingdom e-cigarette sales volume  Time period: 2011 to  Model adjustments:  Cross-price (e- 2014 o Income: yes cigarettes vs.  Design: time series o Cigarette price/tax: yes cigarettes):  Demand data source: o Tobacco policy: no 1.570 store scanner data  Sociodemographics: no  Price type: real  Price/tax data source: store scanner data  Population: consumers Zheng 2016  Country: US  Model type: ordinary least  Own price (e-  Time period: 2009 to squares (OLS) regression cigarettes): 2013  Unit of consumption: sales of e- -1.126 (-1.340, -0.912)  Design: time series cigarettes  Demand data source:  Model adjustments:  Cross-price (e- store scanner data o Income: yes cigarettes vs.  Price type: nominal o Cigarette price/tax: yes cigarettes):  Price/tax data source: o Tobacco policy: no 0.000 (0.000, 0.001) store scanner data  Sociodemographics: yes  Population: consumers Zheng 2017  Country: US  Model type: Almost Ideal  Own price (e-  Time period: 2009 to Demand System cigarettes): 2013  Unit of consumption: sales of -2.054  Design: time series cigars  Demand data source:  Model adjustments:  Cross-price (e- retail scanner data o Income: yes cigarettes vs.  Price type: nominal o Cigarette price/tax: yes cigarettes):  Price/tax data source: o Tobacco policy: no 1.814 retail scanner data  Sociodemographics: yes  Population: consumers Sahadewo  Country: Indonesia  Model type: logistic regression  Own price (kreteks): 2017  Time period: 2015  Unit of consumption: -0.416 (-0.417, -0.415)  Design: cross- participation of kretek use sectional  Model adjustments:  Demand data source: o Income: yes National o Cigarette price/tax: yes Socioeconomic o Tobacco policy: no Survey  Sociodemographics: yes  Price type: tax  Price/tax data source: unclear  Population: unclear Therapeutic Tauras 2003  Country: US  Model type: ordinary least  Own price (Brand 1,  Time period: 1996 to squares (OLS) regression Str 1, Cnt 1): -2.670 1999  Unit of consumption: quantity of  Own price (Brand 1,  Design: time series NRT consumed Str 2, Cnt 1): -2.410  Demand data source:  Model adjustments:  Own price (Brand 1, store scanner data o Income: no Str 3, Cnt 1): -1.640  Price type: real o Cigarette price/tax: yes  Own price (Brand 1,  Price/tax data source: o Tobacco policy: no Str 1, Cnt 2): -2.590 store scanner data o Sociodemographics: no  Own price (Brand 2,  Population: Str 1, Cnt 1): -2.660 consumers  Own price (Brand 2, Str 2, Cnt 1): -2.410  Own price (Brand 2, Str 3, Cnt 1): -2.970  Own price (Brand 2, Str 1, Cnt 2): -1.790  Cross-price (Brand 1, Str 1, Cnt 1, vs. cigarettes): 0.799  Cross-price (Brand 1, Str 1, Cnt 2, vs. cigarettes): 0.744  Cross-price (Brand 2, Str 1, Cnt 1, vs. cigarettes): 0.602  Cross-price (Brand 2, Str 2, Cnt 1, vs. cigarettes): 0.886  Cross-price (Brand 2, Str 3, Cnt 1, vs. cigarettes): 0.621  Cross-price (Brand 2, Str 1, Cnt 2, vs. cigarettes): 0.948 Pipe tobacco Pekurinen  Country: Finland  Model type: ordinary least  Own price (pipe 1989  Time period: 1960 to squares (OLS) regression tobacco): 1987  Unit of consumption: grams of -0.599 (-0.766, -0.432)  Design: time series pipe tobacco consumed  Demand data source:  Model adjustments:  Cross-price (pipe Central Statistical o Income: yes tobacco vs. cigarettes): Office o Cigarette price/tax: yes 2.144 (1.874, 2.414)  Price type: real o Tobacco policy: yes  Price/tax data source: o Sociodemographics: no Central Statistical Office  Population: 15 years+ Roll your own Cornelsen  Country: Ireland  Model type: seemingly unrelated  Own price (roll your 2014  Time period: 1978 to regression (SUR) own): 2011  Unit of consumption: -0.453 (-0.957, 0.051)  Design: time series consumption of roll your own  Demand data source: tobacco  Cross-price (roll your domestic duty-paid  Model adjustments: own vs. cigarettes): consumption o Income: yes 0.074 (-1.196, 1.344)  Price type: real o Cigarette price/tax: yes  Price/tax data source: o Tobacco policy: yes survey of cross- o Sociodemographics: no border prices for 2007- 2011Population: 15 years+ Da Pra 2009  Country: US  Model type: seemingly unrelated  Own price (roll your  Time period: 2006 to regression (SUR) own): 2008  Unit of consumption: sales of roll -0.610  Design: time series your own tobacco  Demand data source:  Model adjustments:  Cross-price (roll your store scanner data o Income: no own vs. cigarettes):  Price data: nominal o Cigarette price/tax: yes 1.600  Price/tax data source: o Tobacco policy: no store scanner data o Sociodemographics: no  Population: not reported Mindell  Country: the  Model type: linear regression  Own price (roll your 2000 Netherlands  Unit of consumption: sales of roll own, time 1):  Time period: 1970 to your own cigarettes -0.910 (-1.300, -0.560) 1995  Model adjustments:  Design: time series o Income: no  Own price (roll your  Demand data source: o Cigarette price/tax: no own, time 2): Dutch Customs and o Tobacco policy: no -0.880 (-0.330, 0.143) Excise Office o Sociodemographics: no  Price type: real  Cross-price (roll your  Price/tax data source: own, time 1): STIROVO (the -1.000 (-1.400, -0.630) Netherlands Foundation on Health  Cross-price (roll your and ) own, time 2):  Population: 15 years+ -1.260 (-2.000, -0.500) Tait 2013  Country: New  Model type: seemingly unrelated  Own price (roll your Zealand regression (SUR) own):  Time period: 1991 to  Unit of consumption: -0.441 (-0.872, -0.010) 2011 consumption of roll your own  Design: time series tobacco  Cross-price (roll your  Demand data source:  Model adjustments: own vs. cigarettes): duty paid monthly o Income: yes 0.867 (0.136, 1.598) tonnes of RYO loose o Cigarette price/tax: yes tobacco released for o Tobacco policy: no sale o Sociodemographics: no  Price type: real  Price/tax data source: unclear  Population: 15 years+ White 2015  Country: Thailand  Model type: multi-nomial  Own price (roll your  Time period: 2005 to regression own): 2006  Unit of consumption: intensity of -0.037 (-0.390, 0.316)  Design: cross- use sectional  Model adjustments:  Demand data source: o Income: yes International Tobacco o Cigarette price/tax: yes Control Southeast o Tobacco policy: yes Asia Survey o Sociodemographics: yes  Price type: real  Price/tax data source: International Southeast Asia Survey  Population: adults Zheng 2017  Country: US  Model type: Almost Ideal  Own price (roll your  Time period: 2009 to Demand System own): 2013  Unit of consumption: sales of -1.678  Design: time series cigars  Demand data source:  Model adjustments:  Cross-price (roll your retail scanner data o Income: yes own vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes -0.609  Price/tax data source: o Tobacco policy: no retail scanner data  Sociodemographics: yes  Population: consumers Smokeless tobacco Bask 2003  Country: Sweden  Model type: seemingly unrelated  Own price (snuff):  Time period: 1964 to regression (SUR) -0.070 1997  Unit of consumption: kilograms  Design: time series of snuff consumed  Cross-price (snuff vs.  Demand data source:  Model adjustments: cigarettes): o Income: no -0.340  Price type: real o Cigarette price/tax: yes  Price/tax data source: o Tobacco policy: no Swedish o Sociodemographics: no MatchPopulation: 15 years+ Chaloupka  Country: US  Model type: ordered probit  Own price (smokeless, 1996  Time period: 1992 to regression and ordinary least current use): 1994 squares (OLS) regression -0.300  Design: repeat cross-  Unit of consumption: current use sectional (participation), intensity of use,  Own price (smokeless,  Demand data source: and monthly smokeless tobacco intensity of use): Monitoring the Future use -0.162 Project  Model adjustments:  Price type: nominal o Income: yes  Own price (smokeless,  Price/tax data source: o Cigarette price/tax: no monthly use): Tax Burden on o Tobacco policy: yes -0.462 Tobacco, the Tobacco o Sociodemographics: yes InstitutePopulation: male adolescents (N=19,581) Ciccarelli  Country: Italy  Model type: generalised methods  Own price (snuff): 2014  Time period: 1871 to of moments (GMM) -0.021 (-0.041, -0.001) 1913  Unit of consumption:  Design: time series consumption of snuff  Demand data source:  Model adjustments: annual budget reports o Income: yes of tobacco industry o Cigarette price/tax: yes management o Tobacco policy: no companies o Sociodemographics: yes  Price type: real  Price/tax data source: annual budget reports of tobacco industry management companiesPopulation: consumers Cotti 2015  Country: US  Model type: ordinary least  Own price (snuff):  Time period: 2004 to squares (OLS) regression -0.053 2012  Unit of consumption:  Design: time series consumption of smokeless  Own price (chewing  Demand data source: tobacco tobacco): store scanner data  Model adjustments: -0.197  Price type: tax o Income: yes  Price/tax data source: o Cigarette price/tax: yes  Cross price (snuff vs. Centers for Disease o Tobacco policy: yes cigarettes): Control and o Sociodemographics: yes 0.744 Prevention and the Office on Smoking  Cross price (smokeless and tobacco vs. cigarettes): HealthPopulation: 1.433 consumers Dave 2013  Country: US  Model type: probit methods  Own price  Time period: 2003 to  Unit of consumption: (smokeless): 2009 participation of smokeless -0.380  Design: repeat cross- tobacco use sectional  Model adjustments:  Cross-price  Demand data source: o Income: yes (cigarettes): National Consumer o Cigarette price/tax: yes -0.770 Survey (NCS) o Tobacco policy: yes  Price type: nominal o Sociodemographics: yes  Price/tax data source:  Population: adults Da Pra 2009  Country: US  Model type: seemingly unrelated  Own price (chewing  Time period: 2006 to regression (SUR) tobacco): 2008  Unit of consumption: sales of -0.210  Design: time series chewing tobacco  Demand data source:  Model adjustments:  Cross-price (chewing store scanner data o Income: no tobacco vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes -0.320  Price/tax data source: o Tobacco policy: no store scanner data o Sociodemographics: no  Population: consumers Joseph 2014  Country: India  Model type: probit methods  Own price (gutkha):  Time period: 1999 to  Unit of consumption: current -0.589 2004 gutkha use  Design: cross-  Model adjustments: sectional o Income: yes  Demand data source: o Cigarette price/tax: yes Global Youth o Tobacco policy: no Tobacco Survey o Sociodemographics: yes  Price type: nominal  Price/tax data source: Global Youth Tobacco Survey  Population: 13 to 15 years old Kostova  Country: India  Model type: logistic regression  Own price 2015  Time period: 2009 and ordinary least squares (OLS) (participation, khaini):  Design: cross- regression -0.091 sectional  Unit of consumption:  Demand data source: participation and intensity of  Own price (intensity, Global Adult khaini use khaini): Tobacco Survey  Model adjustments: -0.069  Price type: nominal o Income: yes  Price/tax data source: o Cigarette price/tax: yes  Cross price Global Adult o Tobacco policy: no (participation, khaini): Tobacco Survey o Sociodemographics: yes 0.018  Population: adults  Cross price (intensity, khaini): -0.035 Nargis 2014  Country: Bangladesh  Model type: logistic regression  Own price (zarda, low  Time period: 2011 to and ordinary least squares (OLS) price): 2012  Unit of consumption: 0.640  Design: cross- participation and intensity of sectional zarda use  Own price (zarda, high  Demand data source:  Model adjustments: price): International Tobacco o Income: yes -0.390 Control Survey o Cigarette price/tax: yes Bangladesh o Tobacco policy: no  Cross-price (zarda vs.  Price type: nominal o Sociodemographics: yes cigarettes):  Price/tax data source: 0.350 self-reported by retailers from stores in the locality  Population: adults Nguyen  Country: Sweden  Model type: Engle-Granger two-  Own price (snus): 2012  Time period: 1955 to step ordinary least squares (OLS) -0.240 (-0.467, -0.013) 2009  Unit of consumption: grams of  Design: time series snus  Cross-price (snus vs.  Demand data source:  Model adjustments: cigarettes): Statistics Sweden o Income: yes -0.193 (-0.376, -0.010) (SCB) o Cigarette price/tax: no  Price type: real o Tobacco policy: yes  Price/tax data source: o Sociodemographics: no Statistics Sweden (SCB)  Population: 15 years+ Ohsfeldt  Country: US  Model type: multivariable  Own price (snuff): 1994  Time period: 1985 logistic regression -0.410  Design: cross-  Unit of consumption: ever sectional smokeless tobacco use  Own price (chewing  Demand data source: (participation) tobacco): Current Population  Model adjustments: -0.630 Survey (N>100,000) o Income: yes  Price type: tax o Cigarette price/tax: yes  Cross price (snuff vs.  Price/tax data source: o Tobacco policy: yes cigarettes): Tax Burden on o Sociodemographics: yes 0.390 Tobacco, the Tobacco Institute  Cross price (chewing  Population: males tobacco vs. cigarettes): aged 16+ 0.490 Ohsfeldt  Country: US  Model type: multivariable  Own price (snuff): 1997  Time period: 1985 logistic regression -0.270  Design: cross-  Unit of consumption: current sectional smokeless tobacco use  Own price (chewing  Demand data source: (participation) tobacco): Current Population  Model adjustments: -0.130 Survey (N>100,000) o Income: yes  Price type: tax o Cigarette price/tax: yes  Cross price (snuff vs.  Price/tax data source: o Tobacco policy: yes cigarettes): Tax Burden on o Sociodemographics: yes 0.130 Tobacco, the Tobacco Institute  Cross price (chewing  Population: males tobacco vs. cigarettes): aged 16+ 0.090 Ohsfeldt  Country: US  Model type: multivariable  Own price (snuff): 1998  Time period: 1992 to logistic regression -0.010 1993  Unit of consumption: current  Design: repeat cross- snuff use (participation)  Cross-price (snuff vs. sectional  Model adjustments: cigarettes):  Demand data source: o Income: yes 0.980 Current Population o Cigarette price/tax: yes Surveys (N=165,653) o Tobacco policy: yes  Price type: tax o Sociodemographics: yes  Price/tax data source: Tax Burden on Tobacco, the Tobacco Institute  Population: males aged 16+ Tauras  Country: US  Model type: probit and  Own price (smokeless, 2007  Time period: 1995 to generalised linear model current use): 2001  Unit of consumption: current -0.121  Design: repeat cross- smokeless tobacco use sectional (participation) and intensity of  Own price (smokeless,  Demand data source: smokeless tobacco use intensity of use): National Youth Risk  Model adjustments: -0.044 Behaviour Surveys o Income: no (YRBS) (N=25,155) o Cigarette price/tax: yes  Cross price  Price type: tax o Tobacco policy: yes (smokeless, current  Price/tax data source: o Sociodemographics: yes use vs. cigarettes): Tax Burden on -0.715 Tobacco, the Tobacco Institute  Cross price (smokeless  Population: male high intensity of use vs. school students cigarettes): -0.413

Zheng 2016  Country: US  Model type: ordinary least  Own price (smokeless  Time period: 2009 to squares (OLS) regression tobacco): 2013  Unit of consumption: sales of -0.405 (-0.574, -0.236)  Design: time series smokeless tobacco  Demand data source:  Model adjustments:  Cross-price store scanner data o Income: yes (smokeless tobacco vs.  Price type: nominal o Cigarette price/tax: yes cigarettes):  Price/tax data source: o Tobacco policy: no -0.931 (-1.837, -0.025) store scanner data o Sociodemographics: yes  Population: consumers Zheng 2017  Country: US  Model type: Almost Ideal  Own price  Time period: 2009 to Demand System (smokeless): 2013  Unit of consumption: sales of -0.532  Design: time series cigars  Demand data source:  Model adjustments:  Cross-price (roll your retail scanner data o Income: yes own vs. cigarettes):  Price type: nominal o Cigarette price/tax: yes -0.442  Price/tax data source: o Tobacco policy: no retail scanner data  Sociodemographics: yes  Population: consumers Waterpipe tobacco Salti 2013  Country: Lebanon  Model type: Almost Ideal  Own price  Time period: 2005 Demand System (AIDS) (waterpipe):  Design: cross-  Unit of consumption: -1.450 (-1.464, -1.436) sectional consumption of waterpipe  Demand data source: tobacco (at home only)  Cross price (waterpipe Household Living  Model adjustments: vs. local cigarettes): Conditions Survey o Income: no -0.370 (-0.476, -0.264)  Price type: nominal o Cigarette price/tax: yes  Price/tax data source: o Tobacco policy: no  Cross price (waterpipe Household Living o Sociodemographics: no vs. imported Conditions Survey cigarettes):  Population: 15 years+ 0.150 (0.132, 0.168) *** denotes P<0.01. **denotes P<0.05. *denotes P<0.1

Appendix 3. Methodology for selecting best estimate of elasticity of demand and for constructing 95% confidence intervals

Author Selection of best estimate of elasticity of Construction of 95% confidence and year demand intervals Bask 2003 Best estimate of elasticity taken from the SUR No measure of precision reported. model, which has higher explanatory power (R2) than the GMM model (see Table 5). Chaloupka Best estimate of elasticity taken from Model No measure of precision reported. 1996 4, which adjusted for more variables than Models 1-3 (see Table 5) Ciccarelli Best estimate of elasticity taken from GMM Standard errors provided (see Table 3) 2014 model after discussion with corresponding to calculate t-statistics which was used author (see Table 3). in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Cornelsen Best estimate of elasticity taken from the Standard errors provided (see Table 1) 2014 Myopic Addiction Model, which has higher to calculate t-statistics which was used explanatory power (R2) than other models in the formula: Confidence intervals = (see Table 1). elasticity of demand*(1±(1.96/t- statistic)). Cotti 2015 Best estimate of elasticity taken the sample No measure of precision reported. not restricted to ever cigarette purchase, after discussion with the corresponding author (see Table 4). Dave 2013 Only one estimate provided by the authors No measure of precision reported. (see text). Da Pra Best estimate of elasticity taken from the Top No measure of precision reported. 2009 Stage Model (long run elasticities), (see Table 5). Escario Only one estimate provided by the authors T-statistics provided (see Table 4), 2004 (see Table 4). used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t-statistic)). Gammon Best estimate of elasticity taken from the main Standard errors provided (see Table 2) 2015 model, which has higher explanatory power to calculate t-statistics which was used (R2) than the alternate model (see Table 2). in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Guindon Best estimate taken from the model where Standard errors provided (see Table 6) 2011 unit values were averaged by cluster over all to calculate t-statistics which was used households, which provided more precise in the formula: Confidence intervals = estimates than the model where unit values elasticity of demand*(1±(1.96/t- were averaged by cluster only over statistic)). households under examination (see Table 6). Huang Best estimate taken from the model with the Standard errors provided (see Table 2) 2014 higher explanatory power (R2) (Table 2) to calculate t-statistics which was used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). John 2008 Best estimate of elasticity taken from Standard errors provided (see Table 5) symmetry constrained model, which provided to calculate t-statistics which was used more precise estimates than the unconstrained in the formula: Confidence intervals = model (see Table 5). Rural and urban elasticity of demand*(1±(1.96/t- estimates pooled in random-effects meta- statistic)). analysis. Joseph Only one estimate provided by the authors No measure of precision reported. 2014 (see Table 5). Kostova Best estimate of elasticity taken from baseline No measure of precision reported. 2015 model, which provided more precise estimates than the reduced model (see Tables 5 and 7). Lee 2005 Only one estimate provided by the authors T-statistics provided (see Table 3), (see Table 3). used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t-statistic)). Mindell Best estimate of elasticity taken from 95% confidence intervals provided by 2000 univariate model, as multi-variate estimates authors (see Table 1). were not available (see Table 1). Estimates from both time periods pooled in random- effects meta-analysis. Nargis Only one estimate provided by the authors No measure of precision reported for 2010 (see Table 2). two of three estimates (see Table 2). Nargis Best estimate of elasticity taken from Model No measure of precision reported. 2014 2, which adjusted for more variables than Model 1 (see Table 1). Nguyen Best estimate of elasticity taken from the error T-statistics provided (see Table 29), 2012 correction model after discussion with used in the formula: Confidence corresponding author (see Table 29). intervals = elasticity of demand*(1±(1.96/t-statistic)). Ohsfeldt Best estimate of elasticity taken from Model No measure of precision reported. 1994 A, which adjusted for more variables than Model B (see Table 2). Ohsfeldt Only one estimate provided by the authors No measure of precision reported. 1997 (see Table 2). Ohsfeldt Only one estimate provided by the authors No measure of precision reported. 1998 (see Table 1.3). Pekurinen Best estimate of elasticity provided by authors T-statistics provided (see Table 2), 1989 based on explanatory power (high R2) and used in the formula: Confidence post-estimation forecast accuracy (low U) (see intervals = elasticity of Table 2). demand*(1±(1.96/t-statistic)). Pesko 2017 Best estimate taken from model in Table 3, 95% confidence intervals provided by which adjusted for more variables than Table the authors (see Table 3). 4. Ringel Only one estimate provided by the authors No measure of precision reported. 2005 (see Table 2). Sahadewo Only one estimate provided by the authors Standard error provided (see Table 3) 2017 (see Table 18). to calculate t-statistics which was used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Salti 2013 Only one estimate provided by the authors Standard deviation provided (see Table (see Table 2). 2) to calculate t-statistics which was used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t-statistic)). Selvaraj Only one estimate provided by the authors Standard error provided (see Table 3) 2015 (see Table 3). Low, middle and high income to calculate t-statistics which was used estimates pooled in random-effects meta- in the formula: Confidence intervals = analysis. elasticity of demand*(1±(1.96/t- statistic)). Shang Best estimate of elasticity taken from the TCP Standard error provided (see Table 2) 2017 survey, based on better precision (see Table to calculate t-statistics which was used 2). in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Stoklosa Best estimate of elasticity taken from No measure of precision reported. 2016 specification 5 (own price) and specification 6 (cross-price), based on a higher explanatory power (high R2) (see Table 2). Tait 2013 Only one estimate provided by the authors Standard error provided (see Table 2) (see Table 2). to calculate t-statistics which was used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Taurus Only one estimate provided by the authors for No measure of precision reported. 2003 each brand, strength and cut. Unable to combine in meta-analysis. See Tables 1 and 2. Taurus Best estimate of elasticity taken from Model No measure of precision reported. 2007 2, which adjusted for more variables than Model 1 (see Table 2). White 2015 Best estimate of elasticity taken from Model 6 Standard error provided (see Table 2) after discussion with corresponding author. to calculate t-statistics which was used in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Zheng Best estimate of elasticity taken from static Standard error provided (see Table 4) 2016 model, which provided more precise estimates to calculate t-statistics which was used than the dynamic model (see Table 4). in the formula: Confidence intervals = elasticity of demand*(1±(1.96/t- statistic)). Zheng Best estimate of elasticity taken from the No measure of precision reported. 2017 unconditional model from the two stages of AIDS model (see Table 5), as these were the only estimates reported in the abstract.