The effect of health information on smoking intensity: does addiction matter?

ABSTRACT

We investigate how health information, such as a notification of hypertension, influences smoking intensity differently among smokers with different levels of addiction. To circumvent the endogeneity of health information, we employ a sharp regression discontinuity design that exploits the discontinuity around the cut-off point for a hypertension diagnosis. The addiction levels are conjectured by the age of smoking initiation. Using individual-level data from China, our results demonstrate that a hypertension notification reduces daily cigarette smoking by 8.01 cigarettes among less-addicted smokers in the short term, while the influence is insignificant among more-addicted smokers; the observed difference is better explained by addiction levels than by health attitudes. The long-term effects of a hypertension notification are insignificant, regardless of addiction levels. Our results may provide new support for the importance of preventing youth smoking and providing regular medical check-ups.

KEYWORDS:

• Health information
• smoking
• hypertension
• addiction
• China

JEL CLASSIFICATION:

• D18
• I12
• I18

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Following the WHO definition, an individual is diagnosed with hypertension when the systolic blood pressure is equal to or higher than 140 mmHg or the diastolic blood pressure is equal to or higher than 90 mmHg.

² We use the terminology ‘hypertension notification’ instead of ‘hypertension diagnosis’ as our treatment, because the intervention is information disseminated by a survey investigator (although the physical examination part was conducted by a physician, nurse, health worker or other health professional), rather than a formal physician diagnosis.

³ In the long-term effect analysis, we use the hypertension notification in previous wave as the key independent variable and the smoking intensity in current wave as the dependent variable, hence, regarding the consecutive waves of 1991 and 1993, the smoking intensity in 1993 is used as the outcome variable while smoking intensity in 1991 is not utilized. To be consistent, we do not include wave 1991 in our short-term effect analysis.

⁴ Among people who have the records about two survey dates (i.e. the dates for the physical examination and the interview) in the CHNS, 38.5% of them had their physical examinations and other interviews on the same date, 7.91% of them had their physical examinations before the interviews, and 53.60% of them had their physical examinations after the interviews. We could not find any geographic or socioeconomic factors that explain the order of the two surveys, and people who had the physical examination before the interview scatter across different regions and socioeconomic status.

⁵ To ensure that the health information shock received from the physical examination is new to the individuals.

⁶ We focus only on SBP because SBP usually attracts more attention in practice (Zhao, Konishi, and Glewwe 2013), thus, those who were diagnosed with hypertension based on only DBP were excluded.

⁷ This condition helps us to disentangle the long-term effect from the short-term effect.

⁸ We also report the smoking intensity and SBP in two consecutive periods in Appendix B table B.2.

⁹ The panel data from the CHNS provides up to seven observations for each individual, which is not enough to obtain a reliable estimate once we include fundamental control variables (e.g. gender, age, and income).

¹⁰ https://digital.nhs.uk/data-and-information/publications/statistical/statistics-on-smoking.

¹¹ https://www.aihw.gov.au/reports/illicit-use-of-drugs/ndshs-2016-key-findings/contents/summary.

¹² Although we also used daily cigarette consumption in the previous wave to define the more-addicted and the less-addicted group in the current wave, the sample size became less than one half and we found no significant effect of the health information.

¹³ Although the results are suppressed from this paper, we also examined the effects of a hypertension notification on smoking cessation, and we could not find any significant effect for both the short-term and the long-term effects. The full estimation results are available upon request.

¹⁴ In the CHNS, most smokers started smoking during this range of ages (see Figure A3 in Appendix A).

¹⁵ We find that our main results are robust to other forms of parametric regressions that include linear and different polynomials with and without interactions (Appendix B, Table B.4).