Table 1

Characteristics of included studies: methods and data

StudyCountryJournalMethodsData
Douglas, Hariharan (1994)10United StatesJournal of Health EconomicsSplit population duration models (probability of ever starting: probit; duration: log-logistic)
Number of time period modelled: 26		Type: retrospective (1954-1979)
Population: representative sample (sample excludes individuals who were > 15 in 1954)
Source: National Health Interview Survey (1978 and 1979)
Sample size: 10,219
Douglas (1998)14United StatesEconomic InquirySplit population duration models (probability of ever starting: ordered probit; duration: log-logistic)
Number of time period modelled: 34		Type: retrospective (1954-1987)
Population: representative sample (sample excludes individuals over 12 in 1954 and individuals who were <25 at time of interview)
Source: National Health Interview Survey: Cancer Risk Factor Supplement (1987)
Sample size: 8745
DeCicca, Kenkel, Mathios (2000)15United StatesJournal of Risk and UncertaintyDiscrete-time hazard models: same as DeCicca, Kenkel, Mathios (2002)Same as DeCicca, Kenkel, Mathios (2002)
Forster, Jones (2001; 2003)16 23BritainJournal of the Royal Statistical SocietySplit population duration models (probability of ever starting: probit; duration: log-logistic). Men and women samples estimated separately.
Number of time period modelled: 64		Type: retrospective (1920-1984)
Population: representative sample
Source: British Health and Lifestyle Survey (1984)
Sample size: 5098 (men 3737; women 4861)
Hammar, Martinsson (2001)17SwedenWorking paperLog-logistic and gamma duration models on subsamples of smokers
Number of time period modelled: 56		Type: retrospective (1945-2000)
Population: Smokers (sample restricted to individuals who were older than 9 and younger than 25 when they started and who were born between 1935 and 1965)
Source: Survey conducted in the counties of Norrbotten and Västerbotten in the northern part of Sweden (2000)
Sample size: 385
Tauras, O'Malley, Johnston (2001)18United StatesNBER working paperDiscrete-time hazard models (model: probit; duration: not reported)
Number of time period modelled: 3 and 4		Type: longitudinal
Population: Eighth and ten graders in 1991, 1992 and 1993 with follow-up surveys at two-year intervals
Source: Monitoring the Future (1991-1999)
Sample size: 8447 (sample size person/years = 15,548)
DeCicca, Kenkel, Mathios (2002)19United StatesJournal of Political EconomyDiscrete-time hazard models (model: probit; duration: unclear, likely non parametric)
Number of time period modelled: 3 (1988, 1990, 1992)		Type: longitudinal
Population: Eighth graders in 1988
Source: National Education Longitudinal Study (1988, 1990, 1992)
Sample size: person/years = 32,392
Ordered probit
Number of time period modelled: 1 (1988-1992)		Sample size: 12,485 (onset between 8th and 10th grade) and 12,262 (onset between 8th and 12th grade)
Time: 1
Ordered probit
Number of time period modelled: 1 (1988-1992)		Sample size: 12,089 (13,989 when imputing missing values)
Time: 1
Probit
Number of time period modelled: 1 (1988-1992)		Sample size: 12,089 (13,989 when imputing missing values)
Time: 1
Glied (2002)20United StatesJournal of Health EconomicsProbit
Number of time period modelled: 1 (see description of dependent variables)		Type: longitudinal
Population: youth aged 14-23 in 1979
Source: National Longitudinal Survey of Youth (1979, 1984, 1992, 1994)
Sample size: 2822 in 1984; 2301 in 1992; 2295 in 1994
López Nicolás (2002)21SpainHealth EconomicsSplit population duration models (probability of ever starting: probit; duration: log-logistic)
Number of time period modelled: 34		Type: retrospective (1957-1990)
Population: representative sample (sample restricted to individuals born after 1957)
Source: Spanish National Health Survey (1993, 1995, 1997)
Sample size: 14,005 (men 7092; women 6913)
Cawley, Markowitz, Tauras (2004)24United StatesJournal of Health EconomicsDiscrete-time hazard models (model: probit; duration: not reported) Models separately estimated for males and females
Number of time period modelled: 3 (1997-2000)		Type: longitudinal
Population: 12-16 years old in 1996
Source: National Longitudinal Survey of Youth 1997 cohort (1998, 1999, 2000)
Sample size: 6255 males and 6027 females
Grignon, Pierrard (2004)25FranceJournal d'Économie Médicale [in French]Split population duration models (probability of ever starting: not reported; duration: log-logistic)
Number of time period modelled: 48		Type: retrospective (1953-2000)
Population: representative sample (sample restricted to individuals aged 20-47 in 2000)
Source: Enquête Santé et Protection Sociale (2000)
Sample size: 9997
Kidd, Hopkins (2004)26AustraliaEconomic RecordSplit population duration models (probability of ever starting: probit; duration: log-logistic)
Number of time period modelled: 38		Type: retrospective (1963-1990)
Population: representative sample (sample restricted to individuals aged 27-37 in 1990)
Source: National Health Interview Survey: Australian Health Survey (1990)
Sample size: 9402 (men 4619; women 4783)
Laxminarayan, Deolalikar (2004)27VietnamHealth EconomicsMultinomial logit
Number of time period modelled: 1 (1992/93-1997/98)		Type: longitudinal
Population:
Source: Vietnam Living Standard Survey (1992/93, 1997/98)
Sample size: 1578
Arzhenovsky (2006)28RussiaQuantile Journal [in Russian]Cox proportional hazards models
Number of time period modelled: unclear		Type: panel 1994-2001 (1994-1996, 1998, 2000-2001)
Population: representative sample (sample restricted to individuals aged 14-65)
Source: Russia Longitudinal Monitoring Survey (1994, 1995, 1996, 1998, 2000 and 2001)
Sample size: 4798 (men 1129; women 3669)
Cawley, Markowitz, Tauras (2006)29United StatesEastern Economic JournalLinear probability; IV (instrument: mother's weight). Models separately estimated for males and females
Number of time period modelled: 7 (1979, 1988, 1990, 1992, 1994, 1996, 1998, 2000)		Type: longitudinal
Population: children born to women aged 14-21 in 1979
Source: Children of the National Longitudinal Survey of Youth, 1979 cohort (1988, 1990, 1992, 1994, 1996, 1998, 2000)
Sample size: 3755 females; 4700 males
Kim, Clark (2006)30United StatesJ Epidemiol Community HealthLogit
Number of time period modelled: 1 (1994/95-2001/02)		Type: longitudinal
Population: representative sample of adolescents, grades 7-10
Source: Add Health, wave 1 (1994/95) and wave 3 (2001/02) (sample restricted to women younger than 18 at wave 1)
Sample size: 2697
Zhang, Cohen, Ferrence, Rehm (2006)31CanadaAmerican Journal of Preventive MedicineLogit
Number of time period modelled: 1 (1994/95-1996/97)		Type: longitudinal
Population: young adults aged 20-24 who did not smoke at baseline
Source: National Population Health Survey (1994/95, 1996/97)
Sample size: 636
Coppejans, Gilleskie, Sieg, Strumpf (2007)32United StatesReview of Economics and StatisticsCox proportional hazards models
Number of time period modelled: unclear		Type: longitudinal
Population: Eighth graders in 1988
Source: National Education Longitudinal Study (1988, 1990, 1992)
Sample size: person/years = 11,146
Grignon (2007)33FranceWorking paperSplit population duration models (probability of ever starting: logit; duration: log-logistic)
Number of time period modelled: not reported		Type: retrospective (years not reported)
Population: unclear
Source: Enquête Santé et Protection Sociale (2004)
Sample size: not reported
Madden (2007)22IrelandApplied EconomicsSplit population duration models (probability of ever starting: probit; duration: log-logistic)
Number of time period modelled: 39		Type: retrospective (1960-1998)
Population: women aged 10-48 in 1998
Source: Saffron Survey (1998)
Sample size: 703 women (sample size person/years = 11,733)
DeCicca, Kenkel, Mathios, Shin, Lim (2008)34United StatesHealth EconomicsDiscrete-time hazard models (model: not reported; duration: not reported)
Number of time period modelled: 4 (1988, 1990, 1992, 2000)		Type: longitudinal
Population: Eighth graders in 1988
Source: National Education Longitudinal Study (1988, 1990, 1992, 2000)
Sample size: person/years = 37,937
DeCicca, Kenkel, and Mathios (2008)37United StatesJournal of Health EconomicsProbit
Number of time period modelled:1 (1992-2000)		Type: longitudinal
Population: Eighth graders in 1988
Source: National Education Longitudinal Study (1992, 2000)
Sample size: 8759
Time: 1
Malhotra, Boudarbat (2009)36CanadaInternational Journal of Economic PerspectivesSplit population duration models (probability of ever starting: probit; duration: log-logistic)
Number of time period modelled: 54		Type: retrospective (1949-2002)
Population: representative sample of population aged 15 + 
Source: Canadian Tobacco Use Monitoring Survey (2002)
Sample size: 22,396
Kenkel, Lillard, Liu (2009)35ChinaHealth EconomicsDiscrete-time hazard models (model: LPM; duration: not reported)
Number of time period modelled: 49		Type: retrospective (1952-2000)
Population: non-representative sample from nine provinces -Guangxi, Guizhou, Heilongjiang,
Henan, Hubei, Hunan, Jiangsu, Liaoning, and Shandong (sample restricted to men aged 21-60 in 2000 and person-year observations corresponding to ages <= 30)
Source: China Health and Nutrition Survey (2000)
Sample size: person/years = 169,386
Liu (2010)38United StatesApplied EconomicsProbit; Probit-IV (instrument: state-level cigarette taxes).
Number of time period modelled: 1 (started to smoke in last 12 months)		Type: retrospective
Population: representative sample of population aged 15 + 
Source: The Tobacco Use Supplement to the Current Population Survey (TUS-CPS) (1992, 1993, 1995, 1996, 1998, 1999, 2001, 2002, 2003)
Sample size: total sample size = 895,668; sample size for initiation specifications not reported
Étilé, Jones (2010)40FranceWorking paperDiscrete-time hazard models (model: LPM (OLS, IV); duration: linear cohort trend, linear time trend, age dummies)
Number of time period modelled: not reported		Type: retrospective
Population: representative sample of population aged 15+ (EPCV), representative sample of population (ES) -sample exclude individuals who were still studying or were under 24 years-old or were 60+ years-old
Source: Enque?te Permanente sur les Conditions de Vie des Me?nages (EPCV) (2001); Enque?te Sante? (ES) (1992, 2003)
Sample size: not reported
Nonnemaker, Farelly (2011)39United StatesJournal of Health EconomicsDiscrete-time hazard models (model: logit; duration: not reported)
Number of time period modelled: not reported		Type: retrospective and longitudinal
Population: 12-16 years old in 1996
Source: National Longitudinal Survey of Youth 1997 cohort (1998-2006)
Sample size: not reported

  • IV, Instrumental Variable; LPM, Linear Probability Model.