The role of time and risk preferences in adherence to physician advice on
health behaviour change
Abstract
Changing physical activity and dietary behaviour in chronic disease patients is associated with
significant health benefits but is difficult to achieve. An often used strategy is for the physician or
other health professional to encourage behaviour changes by providing advice on the health
consequences of such behaviours. However, adherence to advice on health behaviour change
varies across individuals. This paper uses data from a population-based cross-sectional survey of
1,849 individuals with chronic disease to explore whether differences in individuals’ time and risk
preferences can help explain differences in adherence.
Health behaviours are viewed as
investments in health capital within the Grossman model. Physician advice plays a role in the
model in that it improves the understanding of the future health consequences of investments. It
can be hypothesised that the effect of advice on health behaviour will depend on an individuals’
time and risk preference.
Within the survey, which measured a variety of health-related
behaviours and outcomes, including receipt and compliance with advice on dietary and physical
activity changes, time preferences were measured using financial planning horizon, and risk
preferences were measured through a commonly used question which asked respondents to
indicate their willingness to take risks on a 10-point scale. Results suggest that time preferences
play a role in adherence to physical activity advice. While time preferences also play a role in
adherence to dietary advice, this effect is only apparent for males. Risk preferences do not seem
1
to be associated with adherence.
The results suggest that increasing the salience of more
immediate benefits of health behaviour change may improve adherence.
Key words: adherence, time preference, risk preference
2
Introduction
Health behaviours such as physical activity and dietary behaviour are major risk factors for chronic
disease. It is well known that it is challenging to change these behaviours. An often used strategy
is for the physician or other health professional to encourage health behaviour changes by
providing advice on the health consequences of such behaviours. Evidence suggests that physician
advice1 is effective in changing health behaviours (e.g. Loureiro and Nayga, 2007; Kenkel and
Terza, 2001). However, adherence to advice on health behaviour change varies considerably across
individuals (WHO, 2003). To devise incentives for better adherence, it is crucial to gain a better
understanding of the underlying factors that are correlated with whether patients adhere. Many
different determinants of health behaviours and barriers to health behaviour change have been
identified in the literature (Fitzgerald and Spaccarotella, 2009). This paper draws on economic
theory and explores whether differences in individuals’ time and risk preferences can help explain
differences in adherence to physician advice.
A number of papers have highlighted the potential role played by time preference in adherence
(Reach, 2008; Elliott et al, 2008; Christensen-Szalanksi and Northcraft; 1985; Feldman et al,
2002). Adherence to recommended health behaviour changes and medication regimens is an
intertemporal decision with costs and benefits occurring at different points in time. Adherence
generally represents a trade-off between short-term costs and long-term benefits. For example,
adopting smoking cessation advice involves a trade-off between costs now in terms of withdrawal
symptoms in exchange for future health benefits in terms of reduced mortality and morbidity.
Time preferences are key parameters in intertemporal models determining how individuals trade-
1
In the remainder of the paper the term physician advice includes advice from physician or any other health
professional.
3
off outcomes over time. It could therefore be hypothesized that individuals’ time preference would
influence whether they adhere to recommended health behaviour changes. Another feature of the
intertemporal decision is that the future health consequences are uncertain. For example an
individual does not know with certainty whether adopting smoking cessation advice will lead to
future health improvements (they may still experience lung cancer for example). Individuals’
attitude to risk could therefore also play a role in adherence.
The aim of this paper is to investigate the relationship between adherence to physician advice in
health behavior changes and individuals’ time and risk preferences in a sample of chronic disease
patients (heart disease, hypertension, diabetes, stroke). Most previous studies have examined the
correlation between health behaviours (smoking, obesity, excessive alcohol use, drug use) and
time and risk preferences in relatively healthy samples , comparing for example the time and risk
preferences of smokers with those of non-smokers in general population or student samples
(Lawless et al., 2013, Story et al, 2014). The contribution of this paper is two-fold. Firstly, this
study focuses on chronic disease patients rather than general population samples. Understanding
the determinants of health behavior is of particular importance in this group given the larger health
effects associated with these behaviors in this group. Secondly, this study examines whether time
and risk preferences are associated with how individuals adhere to physician advice, that is,
whether time and risk preferences are associated with changes in behavior. Examining adherence
has the potential to inform the development of more effective self-management interventions in
chronic disease. Previous research into the relationship between adherence to physician advice on
health behavior change and time and risk preference is very limited. Chapman et al (2001)
investigated the correlation between time preference and medication adherence in individuals with
4
hypertension and high cholesterol. The results were mixed in that the correlations did not always
have the expected sign and/or were not statistically significant. Brandt and Dickinson (2013)
found a statistically significant relationship between adherence to asthma control medication and
time and risk preferences in a small sample of college students. This paper uses a larger
population-based survey and examines the role of time and risk preferences in people with a range
of chronic conditions (heart disease, hypertension, diabetes, stroke).
Conceptual framework
The conceptual framework draws on the canonical model of demand for health, the Grossman
model. In this model health is viewed as both a consumption good and an investment good.
Individuals can invest in (produce) health through investing resources and time. Choices regarding
level of health investments are made so that the expected discounted utility over the lifetime is
maximised. Grossman did not specifically consider the impact of different time preference rates.
Given that health investments such as physical activity typically involve short-term costs and longterm benefits, it can be deduced that a higher time preference rate will reduce the level of health
investment. Most studies testing this hypothesis empirically have found a significant association
between time preferences and a range of health behaviours (Lawless et al., 2013).
The Grossman model was formulated under conditions of certainty and risk preferences were
therefore not relevant. Dardanoni and Wagstaff (1987) introduced uncertainty into the Grossman
model. Pfeifer (2012) explored the impact of differences in risk preferences within a simplified
Grossman model. Risk preferences affect the subjective expectations of health and productivity
effects of health investments. Individuals who are more risk averse have higher subjective
5
perceptions of health risks and as a result higher expectations of the earning reductions. As a
result, a risk averse individual will make more health investments compared to a risk seeking
individual. Empirical evidence suggests a relationship between risk preferences and health
behaviours (see for example Barsky et al, 1997 and Pfeifer, 2012).
Kenkel and Terza (2001) argue that physician advice enters the model because investment in health
depends on the predicted health consequences. Physician advice will improve the understanding
of the parameters of the health production function. Note that this assumes that the individual will
be receptive to the advice and will accurately interpret the advice. Evidence suggests that most
individuals underestimate the health consequences of unhealthy behaviours in terms of both
probability and magnitude and as a result underestimate the benefits of health investments
(Weinstein, 2000). Provision of information is therefore likely to lead to a higher overall expected
utility of health investments. However, given that health benefits tend to occur relatively far into
the future, the increase in discounted expected utility is lower for individuals with high time
preference rates compared to individuals with a low time preference rate. It is also lower for risk
seeking individuals. The increase in expected utility resulting from a better understanding of the
probabilities of the health consequences is lower in risk seeking individuals as they are less
sensitive to risk compared to risk averse individuals. It is therefore hypothesised that individuals
with high time preference rates and risk seeking individuals are less likely to adhere to physician
advice compared to individuals with low time preference rates and risk averse individuals. It
should be noted that an individual’s perception of the probability and magnitude of the benefits of
health investments may also be a function of their time and risk preferences. For example,
individuals who are risk averse may be more likely to overestimate the risks of unhealthy
6
behaviours. However, this would also result in higher levels of investment. That is, the direct and
this more indirect effect of time and risk preferences on levels of investment work in the same
direction.
Methods
Data
Adherence to physician advice on dietary and physical exercise behaviour changes is modelled as
a function of time and risk preferences. Data were collected as part of a larger cross sectional
survey of chronic disease patients in Western Canada, the Barriers to Care for People with Chronic
Health Conditions (BCPCHC) survey (Weaver et al, 2014). The sampling frame of the BCPCHC
survey consisted of individuals living in Western Canada (Alberta, British Columbia, Manitoba,
Saskatchewan) who were at least 40 years of age and who self-identified as having at least one of
four chronic diseases (hypertension, diabetes, heart disease and stroke) in the 2011 Canadian
Community Health Survey (CCHS). Using these criteria 2,400 respondents were identified for
participation in the BCPCHC, of whom 2,316 were still in scope at the time of survey
administration. The response rate was 80%, yielding a final sample of 1,849 individuals. Data
were collected by telephone using Computer Assisted Telephone Interviews (CATI) by Statistics
Canada. The questions for the survey were designed by subject matter experts in Statistics Canada
Health Analysis Division and project researchers including the authors of this paper. Qualitative
testing of the questionnaire was undertaken in the form of focus group testing. This included
interviews and focus groups with participants having a mix of chronic conditions. With participant
consent, responses from the BCPCHC survey were linked to their corresponding 2011 CCHS
responses to provide more detailed socio-demographic information, including household income.
7
Table 1 shows the descriptive characteristics of the estimation sample. Due to missing data our
sample reduces to 1637 individuals in the case of physical activity advice and 1634 in the case of
dietary advice. Notably income was missing for around 30% of the total sample. For those
individuals, income was imputed using a nearest neighbour technique including postal code
median income, household size and approximate household income range (Yeung and Thomas,
2012).
Adherence variables
Adherence is measured using self-reported adherence questions (Campbell et al 2014).
Respondents were first asked whether they received advice from their doctor or other health
professional. If they stated they had received advice, they were asked whether they had ever
changed their behaviour. If so, they were then asked whether they still did and at what level (none,
some, most or all of the time). Appendix 1 shows the wording of the adherence questions. The
focus of this paper is on the two types of advice that are most often received by all patients in the
sample: advice to change dietary and physical activity behaviours (Campbell et al, 2014). The
dietary advice relates to advice on changing the types of foods individuals eat such as fruits and
vegetables, fish or lean meats, foods high in fibre or foods low in fat to help manage their chronic
disease. The physical activity advice relates to advice on exercising or participating in physical
activities to help manage their chronic disease. Changes in dietary and physical activity behaviours
have been shown to reduce morbidity and mortality (Lichtenstein et al, 2006; Boule et al, 2001;
Blumenthal et al, 2000).
8
The adherence variables are dichotomised into non-adherence (never or no longer adhere). A
binary variable is created that equals one if the individual does not adhere (never or no longer
adhere) and zero otherwise. As very few individuals stated that they no longer adhered these were
merged with those who never adhered. A number of robustness checks are run on the classification
of adherence in the sensitivity analysis.
Self-report measures have been used extensively in the literature as they are practical and
inexpensive but they are subject to social desirability and recall bias (WHO, 2003). The validity
of the measures used in this study is explored by testing whether adherence is associated with Body
Mass Index (BMI), physical activity behaviour and diet quality. These measures were collected
as part of the Canadian Community Health Survey to which the data were linked.
Time and risk preference
In the economics literature risk preferences are generally measured using lotteries and time
preferences using intertemporal trade-offs. Due to constraints on survey length and the mode of
data collection (CATI), it was not possible to use lotteries or intertemporal trade-offs. These
questions tend to be cognitively complex and relatively time consuming.
Time and risk
preferences are therefore measured using proxies in this survey.
Time preferences are measured using financial planning horizon. Planning horizon has been used
in several studies as a proxy for time preference (for example Picone et al, 2004; Khwaja et al,
2007; Samwick, 1998). Adams and Nettle (2009) show that planning horizon and time preference
rate, measured using hypothetical trade-offs over time, are correlated. There are other factors that
9
are likely to be associated with planning horizon such as socio-economic status which are
controlled for in the analysis.
The financial planning horizon question asks respondents:
“In planning your savings and spending which of the following time periods is most important to
you?”
The respondent has the option of choosing: 1) next week; 2) next few months; 3) next year; 4) next
two to four years; 5) next five to ten years; and 6) more than ten years ahead.
Risk preferences are measured through a commonly used risk scale. The scale is used in a number
of household surveys such as the British Household Panel Survey (BHPS). The question asks
respondents to indicate their willingness to take risks on a ten point scale, with one indicating
unwilling to take risks, and ten indicating fully prepared to take risks. The wording of the general
risk question is as follows: “Are you generally a person who is fully prepared to take risks, or do
you try to avoid taking risks?
Dohmen et al (2011) showed that the general risk question
accurately predicts behaviour within incentive compatible lottery experiments.
Other covariates
Other variables of interest in this analysis included chronic condition (categorized as hypertension,
diabetes, heart disease and stroke); gender; age group (categorized as 40-64 years, 65-74 years and
75 years and older); education (dichotomized as less than post-secondary or post secondary);
household income (dichotomized as <$30,000 or $30,000 and greater); marital status
(dichotomized as married/common law or widowed/separated/divorced /single); rural residence
10
(dichotomized as living in a population centre or a rural area); severity (dichotomized as 1 chronic
condition or more than 1 chronic condition); having a usual doctor (dichotomized as yes or no)
and the number of times the respondent saw a family doctor or general practitioner for their chronic
condition in the past year. Respondents were also asked if they had enough information to manage
their chronic condition and were classified as “yes” if they responded that they had all they had
information they needed and “no” otherwise.
Some of the covariates are likely to be correlated with time and risk preferences and including the
covariates may therefore reduce the size and significance of the time and risk preference
coefficients. For example, time preference may affect education which in turn affects nonadherence. However, exclusion of the covariates in the estimation may result in omitted variables
bias in the likelihood of non-adherence. To examine whether time and risk preference are
correlated with the covariates, willingness to take risk and planning horizon were regressed as a
function of the covariates. Appendix 2 shows that risk preferences (willingness to take risk) are
correlated with age, sex, education and not having enough information to manage their condition.
Time preferences (planning horizon) are correlated with age, income and number of times they
saw a family doctor or general practitioner. To understand how the covariates impact the time and
risk preferences coefficients we first estimate a basic model with only time and risk preferences
included and gradually introduce the covariates.
Method of analysis
The main hypotheses will be tested using probit regression analysis. The dependent variable is
non-adherence to advice on health behaviour change (never or no longer adhere). The independent
11
variables are time and risk preferences and the covariates. The covariates consist of factors that
are known to be associated with adherence and include socio-economic factors (gender, age,
education, income, marital status, rural residence), condition related factors (severity – defined as
having more than one chronic condition), and healthcare system factors (whether the individual
has a regular medical doctor and number of times seen or talked to a family doctor or general
practitioner for their chronic condition in the last 12 months) (Kardas et al, 2013). Non-adherence
to advice is modelled separately for the two types of advice, eating certain foods and physical
activity. Whilst the error terms in the two regression models may be correlated as non-adherence
to the two types of advice may be influenced by common unobservable factors such as health
attitude, no efficiency gains are made by using seemingly unrelated regression methods (such as
the bivariate probit) as the regression models have identical independent variables (Greene, 2003).
We model receipt of advice and non-adherence using a probit model which can be derived from a
latent or unobserved variable model. Let y* denote the latent variable such that
𝑦 ∗ = 𝛽0 + ∑𝑘𝑗=1 𝛽𝑥 + 𝜀,
(1)
where 𝜀~𝑁(0,1) and y*=1 if y*>0; 0 otherwise. Then the probability of receipt of advice or nonadherence, y, is given by
𝑃(𝑦 = 1|𝑥) = 𝑃(𝑦 ∗ > 0|𝑥) = Φ(𝛽0 + ∑𝑘𝑗=1 𝑥),
(2)
12
where Φ(𝑧) = ∫ 𝜙(𝑣)𝑑𝑣 is the cumulative normal distribution function with standard normal
density 𝜙(𝑣) = (2𝜋)−1⁄2 𝑒𝑥𝑝(−𝑧 2 ⁄2). This probit model is estimated by maximum likelihood
using the software STATA.
Receipt of advice and non--adherence is modelled as function of time preference (TP), risk
preference (RP) and a range of covariates (D):
𝑅𝑒𝑐𝑒𝑖𝑝𝑡 𝑜𝑓 𝑎𝑑𝑣𝑖𝑐𝑒 = 𝛼0 + 𝛼1 𝑇𝑃 + 𝛼2 𝑅𝑃 + 𝛼3 𝐷 + 𝜀
(3)
𝑁𝑜𝑛 − 𝑎𝑑h𝑒𝑟𝑒𝑛𝑐𝑒 = 𝛾0 + 𝛾1 𝑇𝑃 + 𝛾2 𝑅𝑃 + 𝛾3 𝐷 + 𝜀
(4)
Marginal effects are reported as the parameters α and  will not allow inference of the magnitude
of the impact on the likelihood of receipt of advice and non-adherence.
To explore gender differences, the models are run separately for females and males. The analysis
is also rerun for hypertension patients only. The health consequences of hypertension tend to occur
farther into the future compared to the other chronic diseases (diabetes, stroke and heart disease).
It could therefore be hypothesised that time preferences in particular play a more important role in
adherence to advice on behaviour changes in hypertension patients.
To explore the validity of the self-reported adherence variables, BMI, physical activity and diet
quality are regressed on the adherence measures and a range of covariates (gender, age, education,
income, marital status, rural residence and severity (having more than one chronic condition)).
BMI is measured using self-reported height and weight. Physical activity is measured using a
series of questions about the types and intensity of leisure time activity. Answers to these questions
13
were converted into values for total daily energy expenditure (kcal/kg/day) and further categorized
into three levels (active, moderately active and inactive). For this analysis the variable was
dichotomized into active (incorporating moderately active coded as zero) and inactive (coded as
one). Diet quality is measured using a validated measure based on four questions on fruit and
vegetable intake (Manuel et al., 2012, Manuel 2013). The four dietary variables were combined
into an index based on their individual relationship with mortality that was observed in previous
studies (Manuel et al., 2012; Manuel et al., 2014). A composite score was constructed that
assigned one point for daily fruit and vegetable consumption excluding fruit juice up to a maximum
of eight points. High potato intake (-2 points for ≥7 servings of potato weekly (men) or ≥5
(women) times/week), no carrot intake (-2 points) and high fruit juice intake (-2 points per daily
frequency of fruit juice consumption exceeding once/day up to a maximum of -10 points) were
penalized, as they were shown to be independently associated with mortality. The diet score was
calculated by adding two baseline points to the sum of total points. The resulting score ranged
from 0-10 and for this analysis is coded zero for worst diet (score 0 to 3.5), one for medium diet
(score 3.5 to 6.5) and two for best diet (score 6.5 to 10.0). BMI is modelled using linear regression,
physical activity is modelled using probit regression and diet quality using ordered probit
regression.
Results
Figure 1 shows the distribution of the time and risk preference variables. There is considerable
heterogeneity in time and risk preferences in this sample of chronic disease patients. The
correlation between time and risk preferences is 0.040 (p-value 0.10). Table 2 shows the adherence
variables. The majority of patients have received physician advice (advice from their doctor or
14
health professional) on eating certain foods and physical activity. Non-adherence (never or no
longer adhere) is higher in physical activity (14.6%) compared to eating certain foods (7.6%).
Table 3 shows the marginal effects of the probit regressions for receipt of advice. Older patients
are less likely to have received dietary advice whilst seeing a health professional more often and
disease severity (having more than one chronic condition) increases the probability of receiving
dietary advice. Receipt of physical activity advice is also a function of age and disease severity.
Having a regular doctor increases the probability of receiving physical activity advice whilst
having a household income of less than $30,000 reduces the probability of receiving physical
activity advice.
Table 4 shows the marginal effects of the probit regressions for non-adherence. Risk preferences
are not statistically significant in any of the models. Time preferences (planning horizon) are
associated with non-adherence to physical activity advice. Individuals with a longer planning
horizon (more future oriented) are more likely to adhere (i.e. less likely to not adhere) to advice
on physical activity. The marginal effects show that at the mean value a unit increase in planning
horizon is associated with a 1.9% increase in the probability of adhering to advice on physical
activity. In the case of adherence to advice on eating certain foods, planning horizon is not
statistically significant.
Table 5 shows the full models by gender and for hypertension patients only. Planning horizon has
a consistent effect on non-adherence across the models in the case of physical activity advice.
15
Gender differences are found in the case of adherence to advice on eating certain foods. Planning
horizon is statistically significant in case of males but not for females.
A number of robustness checks on the classification of adherence and the time and risk preference
variables were run. Adhering some of the time was reclassified as non-adherence. Time
preference was no longer significant in this case suggesting that time preference explains the upper
tail of the adherence distribution only. Time and risk preferences were also entered as binary
variables (planning horizon shorter than a year and risk response greater than five) as well as
dummy variables for each of the categories (four in the case of time preference and ten in the case
of risk preferences). Similar results were found across the alternative specifications. It would have
been interesting to explore non-adherence across the two behaviours concurrently but sample sizes
are too small to conduct any robust analyses.
Table 6 shows the results for exploring the validity of the adherence variable. As expected,
individuals who do not adhere to physical activity advice have a higher BMI and are more likely
to report that they are not physically active. Individuals who stated that they did not adhere to
dietary advice are more likely to have a lower diet quality score but no significant relationship was
found with BMI.
Discussion
This paper explored the role of risk and time preferences in non-adherence to physician advice on
health behaviour change in a sample of chronic disease patients. Adherence to dietary and physical
16
activity advice was examined. The results showed that risk preferences were not associated with
adherence in any of the models. Time preferences were associated with adherence to physical
activity advice. The marginal effects were relatively small which is in line with other literature
examining the relationship between time preference and health behaviours (Lawless et al., 2013).
There was a gender effect in case of adherence to dietary advice. Time preferences were associated
with adherence in males only. Jusot and Khlat (2013) found a similar gender difference in the
relationship between time preference and smoking.
There are a number of limitations to this study. Both adherence and time and risk preferences are
likely to have been measured with some error. Adherence to physician advice on health behaviour
changes was self-reported. Evidence suggests that respondents overstate adherence (WHO, 2003).
However, adherence was shown to be correlated with BMI, physical activity behaviour and diet
quality. Whilst these measures were also self-reported they have been shown to predict a range of
objectively measured health outcomes and diseases, including diabetes, stroke, cardiovascular
disease, hospitalization and mortality (Rosella et al. 2011; Manuel et al. 2015; Taljaard et al. 2014;
Manuel et al. 2014; Manuel et al. 2012) The survey only asked whether people were given certain
types of advice and no information was available on the level of detail of this advice, how it was
delivered, how long ago and how often they received advice. We only observed adherence for
those respondents who stated that they received advice from their physician. Individuals may be
less likely to state that they received advice if they did not adhere and this may have introduced a
bias in the adherence analysis. This could be explored in future research by obtaining a more
objective measure of receipt of advice such as a measure based on medical records. Due to the fact
that only a few respondents reported they no longer adhered, it was not possible to examine
17
whether the effect of time and risk preferences is similar for the decision to change behaviour
following physician advice and for the decision to maintain the behaviour changes. This would
be interesting to explore in future research with larger samples. The question also arises whether
planning horizon and the risk scale are valid proxies for time and risk preferences. Empirical
evidence suggests that the proxies are correlated with more robust measures of risk and time
preferences (Dohmen et al, 2011; Adams and Nettle, 2009). More precise measures of time
preferences would have allowed the investigation of alternative discounting models such as the
quasi-hyperbolic discounting model. Evidence suggests that individuals have present bias, that is,
they attach enhanced significance to outcomes that occur in the present (Frederick et al., 2002).
This can explain why individuals often plan behaviour change in the future but fail to follow
through with this. For example, they plan to go to the gym next week but when next week arrives
the temptation of the sofa is too strong. The role of present bias in adherence to physician advice
would therefore be interesting to explore. It is also unclear whether time and risk preferences for
money are the most appropriate type of preferences when the interest is in health investments.
However, time preferences for monetary outcomes tend to be correlated with time preference for
health outcomes (see for example, Lazarro et al, 2001). Moreover the proxies have been shown to
be correlated with health behaviours (see for example Dohmen et al, 2011).
Further research
examining adherence using time and risk preferences for health outcomes is required. Finally, as
the study is based on cross sectional data the correlation between time preference and adherence
does not necessarily reflect causal relationships. Standard economic theory assumes that time
preferences are exogenous and, if this assumption holds, it could be argued that the correlation
reflects a causal effect of time preferences on adherence. However, time preferences may be
endogenous and change over the lifecycle (Becker and Mulligan, 1997).
18
This study provides some support for the role of time preference in adherence to advice on health
behaviour change. This suggests that adherence can be improved by increasing the salience of the
more immediate benefits of health behaviour change. Health behaviour advice often consists of
information on the longer term health benefits. Individuals who are more present oriented, place
relatively less value on longer term benefits than individuals who are more future oriented.
However, health behaviour changes often also have shorter term benefits and increasing the
salience of these more immediate benefits would be effective in particular in individuals who are
present oriented. Increasing the salience of the connection between current activities and future
outcomes may also be effective. For example, Hall and Fong (2003) showed that a ‘time
perspective’ intervention improved physical activity levels in young adults. The intervention
consisted of education and activities to make people more aware of the connections between
present behaviour and future outcomes of physical activity and explained how people often forget
about these connections. Future research could explore whether such an intervention would also
be effective in chronic disease patients.
References
Adams J, Nettle D. 2009. Time perspective, personality and smoking, body mass, and physical
activity: An empirical study. British Journal of Health Psychology 14: 83–105
Becker GS, Mulligan CB. 1997. The endogenous determination of time preference. The
Quarterly Journal of Economics 112; 729-758.
19
Barsky RB., Kimball MS., Juster FT., Shapiro MD. (1997) Preference parameters and
behavioral heterogeneity: an experimental approach in the health and retirement study.
Quarterly Journal of Economics 112; 537-579.
Blumenthal JA, Sherwood A, Gulette EC, Babyak M, Waugh R (2001) Exercise and weight loss
reduce blood pressure in men and women with mild hypertension: effects on cardiovascular,
metabolic, and hemodynamic functioning. Arch Intern Med 160(13): 1947-58.
Boule NG, Haddad E, Kenny GP, Wells GA, Sigal RJ (2001) Effects of exercise on glycemic
control and body mass in type 2 diabetes mellitus: a meta-analysis of controlled clinical trials.
JAMA 286(10): 1218-1227.
Brandt S, Dickinson MS. (2013). Time and risk preferences and the use of asthma controller
medication. Pediatrics 131: e1204-e1210.
Campbell JT, Ronksley PE, Manns BJ, Tonelli M, Sanmartin C, Weaver RG, Hennessy D, KingShier K, Campbell T, Hemmelgarn BR. (2014) The association of income with health
behavior change and disease monitoring among patients with chronic disease. Plos One 9 (4).
Chapman GB, Brewer NT, Coups EJ, Borwlee S, Leventhal H, Leventhal EA. (2001) Value for
the future and preventive behaviour. Journal of Experimental Psychology 7: 235-250.
Christensen-Szalanski JJJ, Nortcraft GB (1985) Patient compliance behaviour: the effect of time
on patients’ values of treatment regimens. Social Science and Medicine 21: 263-273.
Dardoni V, Wagstaff A (1990) Uncertainty and the demand for medical care. Journal of Health
Economics 9: 23-28.
Dohmen T, Falk A, Huffman D, Sunde U,Schupp J, and Wagner GG (2011). Individual Risk
Attitudes: Measurement, Determinants and Behavioral Consequences, Journal of the
European Economic Association 9: 522-550.
20
Elliott RA, Shinogle JA, Peele P, Bhosle M, Hughes DA. (2008) Understanding medication
compliance and persistence from an economics perspective. Value in Health 11: 600-610.
Feldman SR, Chen GJ, Hu, JY, Fleischer AB. (2002) Effects of systematic asymmetric
discounting on physician-patient interactions: a theoretical framework to explain poor
compliance with lifestyle counselling. BMC Medical Informatics and Decision Making 2:8.
Fitzgerald N, Spaccarotella K. (2009). Barriers to a Healthy Lifestyle: From Individuals to
Public Policy—An Ecological Perspective. Journal of Extension 47: 1.
Frederick S, Loewenstein G, O'Donoghue T. (2002). Time Discounting and Time Preference: A
Critical Review. Journal of Economic Literature, 40, pp. 351-401.
Greene. (2003) Econometric Analysis. Prentice Hall.
Hall PA, Fong GT. (2003) The effects of a brief time perspective intervention for increasing
physical activity among young adults. Psychology and Health 18: 685-706.
Heckman, J. (1979). Sample selection bias as a specification error. Econometrica 47 (1): 153–61
Jusot F, Khlat M. (2013) The role of time preferences in smoking inequalities: a populationbased study. Addictive behaviors 38: 2167-2173.
Kardas P, Lewek P, MatyjaszczykM. (2013) Determinants of patient adherence: a review of
systematic reviews. Front Pharmacol. 2013; 4: 91.
Kenkel D, Terza JV. (2001) The effect of physician advice on alcohol consumption: count
regression with an endogenous treatment effect. Journal of Applied Econometrics 16: 165184.
Khwaja a, Silverman D, Sloan F. (2007). Time preference, time discounting and smoking
decisions. Journal of Health Economics 26: 927-949.
21
Lawless L, Drichoutis AC, Nayga RM. (2013) Time preferences and health behaviour: a review.
Agricultural and Food Economics 1: 17.
Lazaro A, Barberan R, Rubio E. (2001) Private and social time preferences for health and
money: an empirical estimation. Health Economics 10: 351-356.
Lichtenstein AH, Appel LJ, Brands M., Carnethon M, Daniels S, (2006) Diet and lifestyle
recommendations revision 2006: a scientific statement from the American Heart Association
Nutrition Committee. Circulation 114(1): 82-96.
Loureiro ML, Nayga RM (2007) Physician’s advice affects adoption of desirable dietary
behaviors. Review of Agricultural Economics 29: 318-330.
Manuel DG, Perez R, Bennett C, Rosella L, Taljaard M, et al. (2012) Seven More Years: The
impact of smoking, alcohol, diet, physical activity and stress on health and life expectancy in
Ontario. Toronto: Institute for Clinical Evaluative Sciences and Public Health Ontario.
Manuel DG (2013) The burden of unhealthy living in ontario: the impact of smoking, alcohol,
diet, physical inactivity and stress on life expectancy. Healthcare quarterly 16(1): 1618.Manuel DG, Perez R, Bennett C, Rosella L, Choi B (2014) 900,000 Days in Hospital: The
Annual Impact of Smoking, Alcohol, Diet, and Physical Activity on Hospital Use in Ontario.
Toronto, ON: Institute for Clinical Evaluative Sciences.
Manuel DG, Tuna M, Perez R, Tanuseputro P, Hennessy D, Bennett C, Rosella L, Sanmarin C,
van Walraven C, Tu JV. Predicting stroke based on health behaviours: Development of the
Stroke Population Risk Tool (SPoRT). PLoS One. 2015. 10(12).
Pfeifer C. (2012) A Note on Smoking Behavior and Health Risk Taking. Nordic Journal of Health
Economics.
22
Picone G, Sloan F, Taylor D. 2004. Effects of Risk and Time Preference and Expected
Longevity on Demand for Medical Tests. Journal of Risk and Uncertainty 28: 39-53.
Reach G. (2008) A novel conceptual framework for understanding the mechanism of adherence
to long term therapies. Patients Preferences and Adherence 2: 7-19.
Rosella LC, Manuel DG, Burchill C, et al. A population-based risk algorithm for the development
of diabetes: development and validation of the Diabetes Population Risk Tool (DPoRT). J
Epidemiol Community Health 2011; 65:613–20.
Samwick AA. 1998. Discount heterogeneity and social security reform. Journal of Development
Economics 57: 117-146.
Story G, Vlaev I, Seymour B, Darzi A, Dolan RJ (2014) Does temporal discounting explain
unhealthy behaviour? A systematic review and reinforcement learning perspective. Frontiers
in Behavioural Neuroscience 76.
Taljaard M, Tuna M, Bennett C, Perez R, Rosella L, Tu JV, Sanmartin , Hennessy D, Tanuseputro
P, Lebenbaum M, Manuel D. Cardiovascualr Disease Population Risk Tool (CVDPoRT):
predictive algorithm for assessing CVD risk in the community setting. A study protocol. BMJ
Open. 2014. Oct 23; 4(10).
Weaver RG, Manns BJ, Tonelli M, Sanmartin C, Campbell DJT, Ronksley PE, Lewanczuk R,
Braun TC, Hennessy D, Hemmelgarn B. (2014) Access to primary care and other health care
use among western Canadians with chronic conditions: a population-based survey. Canadian
Medical Association Journal Open 2(1).
Weinstein ND. (2000). Perceived probability, perceived severity, and health- protective
behavior. Health Psychology 19: 65–74.
23
WHO (2003) Adherence to long-term therapies: evidence for action.
http://apps.who.int/medicinedocs/en/d/Js4883e/
Yeung CW, Thomas S (2012) Income imputation for the Canadian Community Health Survey.
Statistics Canada, Ottawa.
24
Table 1. Descriptive characteristics of the study sample*
All
N
Chronic conditions
Hypertension
Diabetes
Heart disease
Stroke
Sex
Male
Female
Age
40-64 years
65-74 years
75+ years
Education
Post-secondary education
Less than post-secondary education
Income
$30,000+
<$30,000
Marital status
Married/common-law
Widowed/separated/divorced/single
Rurality
Urban
Rural
Severity
One condition
More than one condition
Usual doctor
Yes
No
Enough info to manage condition
Yes
No
%
Males
N
%
Females
N
%
1363
468
381
117
83.3
28.6
23.3
7.2
584
256
195
51
79.2
34.7
26.4
7.3
779
212
186
66
86.7
23.6
20.7
6.9
738
899
45.1
54.9
---
---
---
---
628
509
500
38.4
32.1
30.5
324
221
193
43.9
29.9
26.2
304
288
307
33.8
32.0
34.2
906
732
55.4
44.6
424
314
57.5
42.5
482
417
53.6
46.4
1233
552
69.1
30.9
636
163
79.6
20.4
597
389
60.6
39.5
910
727
55.6
44.4
495
243
67.1
32.9
415
484
46.2
53.8
1228
409
75.1
24.9
536
202
72.6
27.4
692
207
76.9
23.1
1059
579
64.7
35.3
451
287
61.1
38.9
608
291
67.6
32.4
1550
87
94.7
5.3
692
46
93.8
6.2
858
41
95.4
4.6
1126
511
Mean
3.69
68.8
31.2
Range
0-104
496
242
Mean
3.73
67.2
32.8
Range
0-48
630
269
Mean
3.65
70.1
29.9
Range
0-104
Times seen family doctor or general
practitioner for their chronic
condition in the past year
*
estimation sample for physical activity advice. Descriptive statistics for dietary advice sample are
similar and available on request.
25
Table 2. Descriptive statistics advice and adherence variables
Eat certain foods (n=1,634)
Physical activity (n=1,637)
All
Male
Female
All
Male
Female
N (%)
N (%)
N (%)
N (%)
N (%)
N (%)
No
594 (36.3)
251 (34.1)
343 (38.2)
429 (26.2)
188 (25.5)
241 (26.8)
Yes
1040 (63.7)
485 (65.9)
555 (61.8)
1210 (73.8)
550 (74.5)
658 (73.2)
Never or no longer
adhere
79 (7.6)
42 (8.7)
37 (6.7)
177 (14.6)
72 (13.1)
105 (16.0)
Adhere some/most/
all of the time
961 (92.4)
443 (91.3)
518 (93.3)
1031 (85.2)
478 (86.9)
553 (84.0)
Advice
Adherence
26
Table 3. Regression results for receipt of advice
Eating certain foods
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income <$30,000
Married/common law
Rural place of residence
N
Pseudo R2
ME
0.010
0.003
SE
0.007
0.005
ME
0.004
-0.001
-0.069**
-0.191**
0.025
SE
0.008
0.005
0.030
0.030
0.024
ME
0.007
-0.001
-0.083**
-0.216**
0.017
-0.051
0.013**
SE
0.008
0.005
0.030
0.031
0.025
0.056
0.004
-0.005
0.099**
0.026
0.025
1634
0.0011
1634
1634
0.0215
1634
1634
0.0374
1634
ME
0.010
0.011**
SE
ME
0.005
0.009**
-0.095**
-0.187**
-0.009
SE
ME
0.005
0.010**
-0.111**
-0.211**
-0.013
-0.187**
0.000
SE
ME
0.006
-0.001
-0.081**
-0.209**
0.011
-0.053
0.013**
SE
0.008
0.005
0.030
0.032
0.025
0.056
0.004
-0.006
0.099**
0.021
-0.042
0.004
0.026
1634
0.0389
0.026
0.025
0.025
0.031
0.027
0.028
Physical activity
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income<$30,000
Married/common law
Rural place of residence
N
Pseudo R2
*
0.007
0.004
0.007
0.004
0.029
0.029
0.022
0.010
0.086**
1639
0.0051
1639
0.0286
1630
0.0437
0.007
0.004
0.029
0.030
0.022
0.056
0.002
0.023
0.022
ME
0.003
0.008*
-0.105**
-0.188**
-0.023
-0.176**
0.001
0.010
0.089**
-0.036
-0.057**
-0.003
0.008
1639
0.0484
SE
0.007
0.004
0.029
0.031
0.023
0.056
0.002
0.023
0.022
0.023
0.028
0.025
0.025
significant at 10% level; ** significant at 5% level
27
Table 4. Regression results for non-adherence
Eating certain foods
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income <$30,000
Married/common law
Rural place of residence
N
Pseudo R2
ME
-0.008
-0.002
SE
0.005
0.003
ME
-0.006
-0.002
0.026
0.044*
0.025
SE
0.005
0.003
0.021
0.025
0.017
ME
-0.006
-0.002
0.029
0.050**
0.025
0.047
-0.001
SE
0.005
0.003
0.021
0.025
0.017
0.046
0.002
0.013
-0.015
1040
0.0044
1040
0.0145
0.018
0.017
1040
0.0204
ME
-0.006
-0.002
0.026
0.036
0.031*
0.043
-0.001
SE
0.005
0.003
0.021
0.025
0.017
0.045
0.002
0.012
-0.018
0.013
0.029
0.009
0.017
1040
0.0297
0.018
0.016
0.017
0.022
0.018
0.019
Physical activity
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income<$30,000
Married/common law
Rural place of residence
N
Pseudo R2
*
ME
-0.021**
-0.005
SE
0.006
0.004
ME
-0.021**
-0.005
0.005
0.012
-0.021
SE
0.006
0.004
0.024
0.026
0.021
ME
-0.020**
-0.004
-0.003
0.002
-0.028
-0.009
-0.001
0.042*
0.053**
1208
0.0130
1208
0.0144
1208
0.0242
SE
0.006
0.004
0.024
0.025
0.020
0.049
0.003
0.023
0.022
ME
-0.019**
-0.003
-0.003
-0.006
-0.027
-0.014
-0.001
0.042*
0.052**
0.049**
0.000
-0.003
-0.022
1208
0.0303
SE
0.006
0.004
0.024
0.026
0.021
0.047
0.003
0.023
0.022
0.022
0.025
0.022
0.022
significant at 10% level; ** significant at 5% level
28
Table 5. Regression results for non-adherence by gender and for hypertension patients
Eating certain foods
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income<$30,000
Married/common law
Rural place of residence
N
Pseudo R2
Female
ME
SE
0.008
0.006
-0.002
0.004
0.011
0.027
0.051*
0.033
Male
ME
SE
-0.023**
0.007
-0.002
0.005
0.057*
0.033
0.012
0.034
0.091
-0.001
0.080
0.002
0.000
-0.002
0.046
0.004
0.007
-0.018
0.016
0.030
-0.015
0.005
555
0.0397
0.022
0.021
0.022
0.026
0.023
0.024
0.021
-0.023
0.014
0.032
0.035
0.030
485
0.0663
0.026
0.024
0.025
0.036
0.028
0.029
Hypertension
ME
SE
-0.007
0.005
0.002
0.003
0.043*
0.024
0.051**
0.029
0.039**
0.019
0.055
0.051
-0.001
0.002
0.023
-0.021
0.017
0.028
0.011
0.011
859
0.0474
0.019
0.017
0.018
0.024
0.019
0.020
Physical activity
Planning horizon
Willingness to take risk
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage
condition
> 1 chronic condition
Education  post-secondary
Income<$30,000
Married/common law
Rural place of residence
N
Pseudo R2
*
Female
ME
SE
-0.020**
0.009
-0.005
0.006
-0.010
0.034
-0.022
0.036
Male
ME
SE
-0.019**
0.009
0.000
0.006
0.007
0.034
0.021
0.040
-0.025
-0.002
0.070
0.004
0.002
0.000
0.065
0.004
0.049
0.099**
0.034
0.045
-0.018
-0.048
658
0.0454
0.032
0.033
0.030
0.036
0.033
0.030
0.034
0.007
0.063
-0.059
0.008
-0.004
550
0.0283
0.032
0.030
0.031**
0.032
0.032
0.032
Hypertension
ME
SE
-0.018**
0.007
-0.002
0.005
0.021
0.027
-0.001
0.029
-0.011
0.023
0.009
0.055
0.003
0.004
0.046*
0.043*
0.061**
-0.044
0.027
-0.033
1009
0.0340
0.025
0.024
0.024
0.026
0.025
0.024
significant at 10% level; ** significant at 5% level
29
Table 6. Regression results for validity of non-adherence variables
Non-adherence
Age 65-74 years
Age 75+ years
Male
> 1 chronic condition
Education  post-secondary
Income<$30,000
Married/common law
Rural place of residence
Constant
N
Adjusted R2
Pseudo R2
*
significant at 10% level;
Eating certain foods
BMI
Diet quality
Coefficient SE
Coefficient
SE
-0.018 0.610
-0.405** 0.134
-1.612** 0.391
-0.014 0.084
-4.819** 0.443
0.121 0.095
-0.016 0.347
-0.529** 0.075
2.520** 0.344
0.111 0.074
-0.132 0.342
-0.296** 0.073
-0.192 0.423
-0.126 0.091
0.892** 0.375
-0.325** 0.080
0.085 0.383
-0.021 0.081
29.406 0.385
1087
1032
0.1273
0.0460
**
BMI
Coefficient
2.067**
-1.888**
-5.105**
-0.248
2.404**
0.078
-0.228
1.065**
0.089
29.482
1253
0.1475
Physical activity
Physically active
SE
Coefficient SE
0.434
0.872** 0.108
0.368
-0.072 0.086
0.405
0.136 0.095
0.322
-0.136* 0.075
0.322
0.175** 0.075
0.321
0.132* 0.075
0.389
0.195** 0.091
0.346
-0.031 0.081
0.357
0.089 0.083
0.361
-0.308 0.084
1291
0.0627
significant at 5% level
30
Figure 1. Histograms time and risk preferences
Planning horizon (time preferences)
Willingness to take risk (risk preferences)
Unwilling to
take risks
Fully prepared
to take risks
31
Appendix 1. Survey questions
Adherence questions
The next questions are about things that a doctor or other health professional may have suggested
to help manage your ^CONDITION, and the things that people might do as a result of being
diagnosed
Has a doctor or other health professional ever suggested / recommended:
eating certain foods such as fruits and vegetables, fish or lean meats, foods high in fibre or foods
low in fat to help manage your chronic disease?
 Yes
 No
Did you ever change type of foods you eat, for example choosing more fruits and vegetables, fish
or lean meats, foods high in fibre or foods low in fat to help manage your chronic condition?


Yes
No
Are you still doing this:
 All of the time
 Most of the time
 Some of the time
 None of the time
Has a doctor or other health professional ever suggested / recommended:
that you exercise or participate in physical activities to help manage your chronic condition?
 Yes
 No
Did you ever exercise or participate in physical activities to help manage your chronic condition?


Yes
No
32
Are you still doing this




All of the time
Most of the time
Some of the time
None of the time
Planning horizon
In planning your saving and spending, which of the following time periods is most important to
you?

The next week

The next few months

The next year

The next 2 to 4 years

The next 5 to 10 years

More than 10 years ahead
Risk
Are you generally a person who is fully prepared to take risks, or do you try to avoid taking
risks?
On a scale ranging from 1 (Unwilling to take risks) to 10 (Fully prepared to take risks), where
would you place yourself?

1: Unwilling to take risks

2

3

4

5

6
33

7

8

9

10: Fully prepared to take risks.
34
Appendix 2. Regression results for time and risk preferences
Age 65-74 years
Age 75+ years
Male
No usual doctor
Times seen doctor
Not enough info to manage condition
> 1 chronic condition
Education  post-secondary
Income <$30,000
Married/common law
Rural place of residence
Constant
N
Adjusted R2
*
Willingness to take risk
Coefficient
SE
-0.028
0.146
-0.349**
0.155
0.727**
0.125
0.234
0.269
0.005
0.014
-0.331**
0.130
-0.117
0.129
-0.532**
0.125
-0.285*
0.150
0.043
0.134
0.220
0.139
5.413
0.153
1639
0.0502
Planning horizon
Coefficient
-0.162*
-0.341**
0.015
-0.180
-0.021**
-0.040
-0.122
-0.144*
-0.334**
-0.085
0.034
3.941
1639
0.0339
SE
0.095
0.100
0.081
0.175
0.009
0.084
0.084
0.081
0.097
0.087
0.090
0.099
significant at 10% level; ** significant at 5% level
35