Health and Wealth (WP10/05)

Abstract#

This paper analyses the relationship between net wealth and health using Waves 1 to 3 of the Survey of Family, Income and Employment (SoFIE).  The results show that lower net wealth is associated with worse health over a range of differing measures of health. The paper acknowledges but does not attempt to resolve the complex issue of causality; does health cause wealth or vice versa?

Physical and mental wellbeing were both found to be positively associated with net wealth.  These measures of wellbeing were decomposed by the occurrence of a health failure, defined as an injury or illness lasting more than one week.  The results led to further inspection of the characteristics associated with health failures.  This revealed that those who experienced a health failure had, on average, less wealth and worse self-rated health than those who did not.

The progressive nature of poor health and lower net wealth was reinforced by considering self-rated health.  There was a clear negative relationship between poor self-rated health and lower net wealth over the five categories of self-rated health.

A series of chronic health conditions were also examined.  The presence of these conditions was associated with lower net wealth though certain conditions were not always significant.  Other than the presence of depression or schizophrenia, each chronic condition was decomposed by age of diagnosis revealing that asthma is more significant in the short term.  For conditions other than asthma the coefficients were not significantly different.
The analysis of wealth excluded those with zero or negative values for their wealth.  To provide a more complete picture, the probability of having zero or negative net wealth was modelled.  This revealed that individuals reporting poorer health were more likely to have non-positive net wealth.

This study has relied on cross-sectional data from SoFIE.  Once the full eight years of longitudinal data become available, a richer analysis of the impact of changes in health status over time on assets, liabilities and net wealth will be possible.

Acknowledgements#

Many thanks to K Henderson, M Gibbons, H Holt and G Scobie for their input, advice and feedback.

The Health Research Council of New Zealand, and Health Inequalities Research Programme of the University of Otago, Wellington, are acknowledged for funding and establishing the SoFIE-Health data utilised in this publication.

Disclaimer#

The views, opinions, findings and conclusions or recommendations expressed in this Working Paper are strictly those of the author. They do not necessarily reflect the views of the New Zealand Treasury or the New Zealand Government.  The New Zealand Treasury and the New Zealand Government take no responsibility for any errors or omissions in, or for the correctness of, the information contained in this Working Paper. The paper is presented not as policy, but with a view to inform and stimulate wider debate.

Access to the data used in this study was provided by Statistics New Zealand in a secure environment designed to give effect to the confidentiality provisions of the Statistics Act 1975.  The analysis in this paper is based on data from the Survey of Family, Income and Employment (SoFIE).  Statistics New Zealand has initiated a systems review for SoFIE.  Therefore data contained in this paper could be subject to change.  However, any errors in the analysis are those of the author, not Statistics New Zealand.

1  Introduction#

The central question addressed by this paper is: To what extent can variations in net wealth among the New Zealand population, aged 15 years and older, be explained by differences in health status? There has been little study of the association between health and wealth in New Zealand, although strong links between the two have been found overseas.

Wealth can be an indicator of both economic success and of financial security. The opportunity to accumulate wealth will depend, in part, on an individual's participation in the labour force and their productivity, factors which in turn may well be governed by their health status. Accumulated wealth represents an individual's ability to consume if they are unemployed or out of the workforce and personal wealth can be invested to generate income. As the New Zealand population ages, an increasing proportion of the population will be retired and will be drawing on their accumulated wealth for consumption. Imperfect health may lead to less time spent in the workforce and higher health costs.

Health is of interest in New Zealand as the Government allocates about 20% of each year's budget to the provision of public health (Bryant, Teasdale, Tobias, Cheung and McHugh, 2004). In making funding decisions the Government decides how much of the cost of health care should be borne by the public and how much by the consumers of health care. If imperfect health has negative effects on a person's ability to accumulate wealth then finding ways to address this will improve the overall wellbeing of the New Zealand population.

The analysis reported in this paper is based on unit record data from a household survey. Regression models were estimated for net wealth as a function of health and other variables. Poor health may have negative effects on an individual's rate of wealth accumulation, while lower levels of wealth may have negative health implications. Establishing causality was not possible with the available data. The focus is on modelling the association between health and wealth without considering causality. This paper also does not consider the effects of changes in policy on health or wealth.

Health status and wealth both develop gradually over the life cycle. Furthermore, they can both also change rapidly in response to events. However, these sudden events are likely to be correlated with long-run developments: for example, people with more human capital are less likely to be hit with unemployment shocks to wealth; people with a lifetime of poor lifestyle are more likely to experience an adverse health shock, and to have less wealth. This underscores the fact that health and wealth are both jointly determined and evolve over time. Ideally, one needs longitudinal data to adequately capture the dynamic interrelationships. Furthermore, identifying the direction of causality between wealth and health is a complex challenge, and an issue far from resolved in the literature. This paper relies on exploring associations without pretending to establish causality.

This study relies on cross-sectional data and as such does not attempt to explore the long-run evolution and interrelation between health and wealth. To the extent that the study identifies an association between health and wealth from the cross-sectional data, it is recognised that this is merely a snapshot of a process that evolves over the life cycle. One can envisage that some people will place greater emphasis on the future (ie, have lower discount rates) and as a result invest more in both their human and non-human wealth, resulting in jointly determined higher levels of health and wealth.

This paper is set out as follows: Section 2 reviews work already done in similar areas. Section 3 explains the data and Section 4 sets out the models used in this paper. The main results can be found in Section 5, with supporting results in Section 6. Section 7 concludes. An extensive set of results is presented in the appendices.

2  Previous studies#

Net wealth is defined as accumulated savings and asset income, plus inheritance less gifts. Individuals use their income for consumption, for savings and to improve the value of their assets. Wealth is invested to earn a return that compensates the owner for their forgone consumption (Headey, Marks and Wooden, 2005). Life cycle theory suggests wealth will increase over the course of a person's working life because of savings and investment income and will decrease after retirement as they draw down accumulated wealth to sustain consumption.

The longitudinal importance of income and age on net wealth were both identified in Australia using the Household Income and Labour Dynamics in Australia (HILDA) survey. Although wealth and income were well correlated during certain parts of the life cycle, there was an overall low correlation between wealth and income. Once people had accumulated sufficient wealth they could retire early. This was most notable amongst the self-employed (Creedy and Tan, 2007).

Headey and Wooden (2004) defined wellbeing using four different measures: personal utility and satisfaction, mental ill being, financial stress and financial security. HILDA data showed that greater financial stress and lower financial security both corresponded to progressively lower levels of net wealth and lower levels of income. Personal utility and mental ill being were both more significantly correlated with wealth than with income.

Health data from SoFIE has been used to consider labour force participation, with poor health associated with a decrease in the likelihood of labour force participation. Part-time employment is also affected, though to a reduced degree, suggesting that poorer health not only reduces the likelihood of participation but also the number of hours worked by those still participating (Holt, 2010). Both these effects will reduce an individual's income and are expected to decrease their rate of wealth accumulation.

Holt (2010) considered the possibility of “rationalisation endogeneity” biasing the relationship between self-rated health and labour force participation (ie, that a respondent who reports they are not in the labour force may be inclined to rate their health worse than it is to justify their non-participation).

Self-rated health has been shown to deteriorate faster for individuals employed in particular industries, most notably those involving manual work (Case, and Deaton 2003). Differences in self-rated health across the income distribution appear to be owing to the loss of income from health-related absence from the labour force.

Case (2001) also considered the effect of large increases in wealth owing to the introduction of pensions for black South Africans.[1] Owing to law and superannuation changes after the ending of apartheid, many black South Africans now receive superannuation income that is more than double the median income for black South Africans. When retirees live in a household where income is pooled, receiving superannuation appears to result in better health for the entire household. When retirees live in a household where income is not pooled, receiving superannuation was only associated with an improvement in the retiree's health.

While a relationship between health and wealth or health and income can often be detected, the direction of the association is ambiguous. One approach used in the literature to address this problem is through the use of instrumental variables. This technique requires identifying a variable that is associated with net wealth but not correlated with health status. Typically it has not proven easy to identify such a variable.

Meer, Miller and Rosen (2003) proposed the use of inheritance as an instrumental variable. Of 3,302 individuals, observed over two consecutive five-year periods, there were 297 recorded inheritance receipts worth more than $10,000 (this is 4.5% of the observed person time periods). While a strong link between health and wealth was found before the introduction of inheritance, the instrumental variables approach resulted in the wealth coefficients no longer being statistically significant. It was concluded that the relationship between health and wealth was not driven by short-term changes in wealth.

Health has been identified as being associated with the composition and not just the level of assets held. Poor health was shown to be associated with a decrease in the likelihood of having less stable assets and an increase in the proportion of total assets held as “safe assets”. This relationship persisted even when respondents' attitudes to risk, time horizon, bequest motives and health insurance were considered. Proof of causality was not established but was discussed. The hypothesis that investment choices might determine health status was not supported (Wu and Rosen, 2003).

3  The Survey of Family, Income and Employment#

The data for this study comes from the Survey of Family, Income and Employment (SoFIE) conducted by Statistics New Zealand. SoFIE is a national longitudinal survey that commenced in 2002 and will continue until 2010.[2] The target population for SoFIE is the usually resident population of New Zealand, aged 15 years and older, who are living in private dwellings.

Approximately 11,500 households agreed to be interviewed, with data collected from 22,000 individuals aged 15 and over. All individuals who were interviewed in Wave 1 are Original Sample Members (OSMs). Their children under the age of 15 will be interviewed in subsequent waves once they have had their 15th birthday and will also be considered OSMs. Each year all OSMs and anyone else, aged over 15, living with them are interviewed.

Each wave of SoFIE asks about family and personal characteristics, income, education and labour force participation. Waves 2, 4, 6 and 8 contain a module of detailed wealth questions. Waves 3, 5 and 7 contain a module of detailed health questions[3]. This study was undertaken between the releases of Waves 3 and 4.

3.1  Population of Interest#

The population of interest for this study is restricted by data availability. This population is represented by OSMs aged 17 and over in Wave 3 who answered Waves 1, 2 and 3. This sample represents 2.88 million individuals.

Statistics New Zealand provided longitudinal weights for the OSMs that adjusted for non-response and ensured that the SoFIE data matched estimates for age, sex and ethnicity of the New Zealand population as at October 2002. Longitudinal sample weights continue to be assigned for all OSMs, including those who do not respond in subsequent waves.

Although SoFIE was designed to be a random sample of the target population, non-response, subsequent attrition and non-response to particular questions have reduced its representativeness. The response rate for the first wave of SoFIE was 77% while 80% of the OSMs remained in the sample at Wave 3. A small number who responded to SoFIE did not respond to the detailed health and/or wealth modules. Because these respondents were assigned weights, they contribute to estimates of population totals. As a result, the population represented by the sample used in the analysis is reduced by the number of individuals represented by these respondents. However, the use of weights is not guaranteed to eliminate non-response bias. The base sample for the regressions reported in this paper consists of 17,043 respondents, representative of 2.78 million individuals.[4] For the remainder of the analysis, this population will be referred to as the “longitudinal population”.

In order for non-response (owing to initial non-response, attrition and non-response to specific questions) to not bias the estimates of the relationship between health and wealth, it would need to be random with respect to health and wealth. However it is not possible to test this assumption with the available data. Ideally, results from this analysis will be representative of New Zealanders who were within the target population of the survey in 2002, and were still in the target population for Wave 3.

3.2  Measures of wealth#

In Wave 2, respondents are asked to provide the values of their assets and liabilities. This includes property, mortgages, superannuation (other than New Zealand Superannuation), life insurance, trust funds, financial funds, bank assets, bank liabilities, personal loans, student loans, durables, cash, business value, hire-purchase, credit card and overdraft debt. Two non-specific categories were provided for miscellaneous assets and liabilities. These were aggregated into total assets and total liabilities, with net wealth defined as the difference.

Henderson and Scobie (2009) noted that the SoFIE data underestimates the levels of assets and liabilities when compared to data from the Reserve Bank of New Zealand. While there are elements that differ between the two, these are unlikely to explain the difference. SoFIE was not compared to the Reserve Bank of New Zealand data during the course of this study.

3.2.1  Net wealth

We consider net wealth, the dependent variable, as a function of health variables and control variables. Table 1 gives percentiles for net wealth. While net wealth has a wide range, a significant proportion of total wealth is held by the wealthiest individuals. The range between the 95th and 99.9th percentiles is of greater magnitude than that between the 5th and 95th percentiles.[5]

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Taking natural logarithms of net wealth scaled down the large positive outliers. This produced a bell shaped curve with similar sized tails on both sides of the distribution. Regression models using the logarithm of net wealth as the dependent variable gave a significant improvement in goodness-of-fit relative to using actual net wealth.

Under New Zealand Superannuation (NZS), New Zealand citizens and permanent residents receive superannuation payments from the Government if they are aged 65 years or older until the time of their death. The expected future cash flows from NZS represent a significant asset to the majority of the population.[6] The present value of future NZS payments was not included in the measure of total assets for this study. As the value to an individual depends on life expectancy, its inclusion would have necessitated allowing for the effect of an individual's health status on life expectancy. This was not attempted.

The term “wealth” in this paper refers to net wealth (assets less liabilities). So individuals with high levels of total assets, who appear to be “wealthy”, but who have high levels of total liabilities are not necessarily considered wealthy for the purposes of this report.

3.2.2  Student loans and negative net wealth

Approximately 6.5% of the longitudinal population were estimated to have total liabilities in excess of their total assets. Liabilities are often insured against a corresponding asset of equal or greater value. However, our measures of assets and liabilities consider only tangible assets and liabilities.

Many tertiary students receive student loans to fund their education. A student loan enables a person to borrow against their future earnings. Economic theory suggests that education increases a student's human capital and, because of this, their future earnings. Following the work of Henderson and Scobie (2009), the liability of a student loan was exactly offset by the conservative assumption that there would a human capital asset of at least equal value.

Table 2 summarises the percentage of the longitudinal population with negative net wealth. Only 2.5% of the longitudinal population were still estimated to have negative net wealth after the inclusion of an offsetting human capital asset.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

The assumption that investment in education creates an asset of equal value to student loans is in fact extremely conservative. In general, an individual undertaking study would expect the present value of their future earnings to increase by more than the cost of their study. Furthermore, although every individual in the population will possess human capital of some measure, we only include a human capital asset for those individuals with student loans.

When decomposing net wealth into assets and liabilities the human capital asset was included as a negative liability rather than as an asset. So human capital is not included in total assets but instead offsets student loans in the measure of total liabilities.

3.3  Measures of health#

Several measures of health from the SoFIE data were used in order to provide a broader picture of the association between health and wealth. Health measures with a wider focus tend to be more subjective than measures with a narrower focus.

Each wave contains a single question asking respondents how they rate their health (this is referred to as self-rated health). In Wave 3 a more detailed module of health questions is asked. This includes several internationally recognised health surveys: the Short Form Health Survey (SF36) and the Kessler 10-item scale (K10) were both used. Respondents were also asked about the presence of eight chronic conditions.

Self-rated health is the most widely answered measure of wellbeing available with less than 0.1% of respondents failing to rate their own health. For the SF36 and K10 questionnaires 1.4% of respondents did not provide responses, and up to 0.7% of respondents did not provide answers to the questions on chronic conditions.

3.3.1  Short Form Health Survey (SF36)#

The SF36 is a health questionnaire consisting of 36 questions that can be split into eight measures of health. Four of these measures relate to physical health and four relate to mental health. Physical and mental health summary variables are calculated from the health measures.

The SF36 has been tested and found reliable for use in New Zealand (Scott, Sarfati, Tobias and Haslett, 1999), but its validity for Māori or Pacific Island populations was later questioned as they do not view mental and physical health to be as separable as the survey assumes (Scott et al, 2000).[7]

3.3.2  Kessler 10-item scale (K10)#

The K10 is a scale measuring non-specific psychological distress. It consists of 10 questions that seek to measure anxiety, depression and negative emotional states a person may have experienced in the four weeks prior to the interview. Items are rated 1 (none of the time) to 5 (all of the time). Scores for the 10 items are summed yielding a total score between 10 and 50 with lower scores signifying better health.

The K10 has been tested and validated in Australia (Andrews and Slade, 2001). It has seen wide use in Australia and around the world and has also been used in mental health surveys in New Zealand (Carter, Hayward and Richardson, 2008).

3.3.3  Choice of health survey regressors#

The SF36 and K10 measures of wellbeing are referred to as the health survey measures. These surveys were analysed by the Otago School of Medicine, Wellington, who calculated measures of wellness from the raw responses.

Following the design of other analysis (Headey and Wooden, 2004), physical and mental wellbeing were treated separately. This allows for different types of ill health to have differing effects on net wealth. The physical functioning (PF) component of the SF36 survey was chosen as the preferred measure of physical discomfort. The K10 survey was chosen as the preferred measure of psychological distress.

Each of the measures of physical wellbeing collected in the SF36 survey, including the physical component summary, were considered for use in the model. The PF component was selected because of its goodness-of-fit and the nature of the questions that determine this score. The questions asked about the influence of the respondent's health on their ability to perform a range of common physical activities, including walking distances, climbing stairs, bending, lifting, bathing and dressing. The PF measure is scored from 0, representing significant problems, to 100, representing the absence of problems.

The K10 score was used as the measure of mental wellbeing, as opposed to a measure from the SF36, owing to the low correlation with the physical functioning measure and for its improvement to the goodness-of-fit of the model.[8]

3.3.4  The application of SF36 and K10 regressors#

Categorical measures of physical and mental wellbeing were used as the relationship between the log of wealth and wellbeing may not be linear. Responses to each health measure were broken into three categories.

Table 3 gives the bounds on the categories, the percentage of the longitudinal population in each category and the mean and median net wealth in each category.

Source: SoFIE Waves 1-3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note: These results are not corrected for age, and younger respondents are likely to report low net wealth and better than average health status.

The breakdown of K10 scores into categories follows recommended criteria (Diener, Suh, Lucas and Smith, 1999; Phongsavan, Chey, Bauman and Brooks, 2006). This specifies four categories, breaking the high psychological distress category into high (22-29) and very high (30-50). Individuals with K10 scores greater than 30 are expected to meet the criteria for clinical intervention. The proportion of the longitudinal population with estimated K10 scores in excess of 30 (1.6%) was too small for the regression coefficient to be of use. The high and very high psychological distress categories were therefore merged.

A fourth category was also attempted with the PF score, very high physical discomfort, separating high and very high discomfort at a score of 20. Because this created a category with too few observations to provide a useful estimator the very high and high discomfort categories were merged. Only 2.5% of the longitudinal population were estimated to suffer from very high physical discomfort.

Figure 1 – Distributions of K10 and physical functioning scores

 

Source:  SoFIE Waves 1-3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. The three vertical lines in the body of each plot show the cut-off points between the different categories. The outermost line in each plot shows the cut-off between low and moderate discomfort/distress. The middle line in each plot shows the cut-off between moderate and high discomfort/distress. The inner most line in each plot shows the attempted (but not used) cut-off between high and very high discomfort/distress.

Figure 1 shows the distributions of the PF and K10 descriptors. The divisions between the categories for high, moderate and low discomfort and distress are included to show the approximate proportions for the longitudinal population in each category. The divisions of high discomfort and distress into high and very high have also been provided.

3.3.5  Self-rated health#

Respondents were asked in each wave to rate their own health. They had five choices of response: excellent, very good, good, fair and poor. Self-rated health from Wave 2 was used; this is the only measure of health that does not come from Wave 3. This measure is used as it comes from the same wave as the wealth data and because it may include aspects of health not covered by SF36, K10 or the chronic conditions.

Self-rated health was included in the model as a categorical variable. The five possible responses for self-rated health were used as distinct categories. Merging of adjacent categories was considered, but coefficient testing revealed the different self-rated health categories were distinct from each other.

Self-rated health has the potential to be misleading as there may be no standard for responses between individuals. Two respondents with the same level of wellbeing, may rate their own health differently. Headey and Wearing (1992) suggest that people rate their own health in comparison to others of the same gender and age, to their parents and siblings and to their own recent past. If a respondent's perception is affected by non-health-related events then these may also influence their response.

Furthermore, responses may change owing to factors unrelated to ongoing health, such as catching the flu in the last month. The estimated “effect” on wealth may be biased if health is not an exogenous variable (ie, it is endogenous). A variable is endogenous if it is affected by the dependent variable or if there are unobserved variables that affect both variables. A particular type of endogeneity that is more likely to affect the self-rated health regressions than those based on more objective measures is referred to as “rationalisation bias”.[9] It is possible that an individual's wealth influences their current perspective about their health. For example, individuals with low levels of wealth, which may be linked to factors such as being unemployed, may be inclined to understate their health in order to justify their low wealth.

3.3.6  Chronic conditions#

As part of the health module in Wave 3, respondents were asked whether they have ever been diagnosed by a doctor with any of the following conditions:

• asthma
• high blood pressure
• high cholesterol
• heart disease
• diabetes
• a stroke
• migraines
• depression or schizophrenia.

These will be referred to as the chronic conditions.

For each condition, other than depression or schizophrenia, the age of diagnosis was asked. This enabled the chronic conditions to be backdated so only respondents who had been diagnosed with a condition by Wave 2 were recorded as suffering from one. Individuals diagnosed with depression or schizophrenia in Wave 3 were assumed to have been suffering from the condition during Wave 2.

No indication of severity of the condition is asked, nor whether the respondent still suffers from the condition (or has suffered from it recently in the case of a stroke). As a consequence, some respondents who report having been diagnosed with a condition may no longer be affected by it. The chronic condition indicators may therefore not be an accurate indication of the presence of negative health effects owing to these conditions. Alternatively, the indicators may be advantageous as they enable the association between past ill health and current wealth to be considered.

3.4  Strengths and limitations#

Empirical evidence shows that a significant proportion of a country's wealth is held by a minority of the population. This means it is easy to under sample these people. The more uneven the distribution of wealth the more likely it is that the very wealthy will be excluded from the sample (Headey and Wooden, 2004). Furthermore, the wealthier members of the population may have greater incentives to withhold information about the size of their assets. Because of this, the very wealthy are unlikely to be well represented by this study.

There is a one-year difference between the collection of the wealth data, in Wave 2, and the health data, in Wave 3. As the age of diagnosis was asked for the chronic conditions these could be backdated to Wave 2. Self-rated health is asked in every wave, but there is no way to backdate the health measures calculated from the SF36 and K10 health surveys.

It is therefore necessary to assume that reported health from the SF36 and K10 surveys does not differ significantly from one year to the next in order to use them as proxies for health in Wave 2. For many respondents, their actual health in Wave 2 is expected to closely match their recorded health in Wave 3. The health module includes a question as to whether the respondent experienced an illness or injury that hindered their normal activities for seven days or more in the 12 months preceding the interview. For respondents who answered “yes” to this question this assumption is potentially less valid.

The SoFIE survey will collect a vast amount of information over the eight-year period, for which it runs, between 2002 and 2010. Additional waves will enable a more systematic study of the changes in wealth and health over time. Despite the fact that the complete longitudinal data set will only cover eight years, this research will be potentially more powerful than the cross-sectional analysis of the present study. Models of savings over time and comparisons between cross-sections will provide further understanding of how health and wealth change over time.

4  Models#

The following analysis uses four main models - two linear regression models and two logistic models. Results from the linear regression models can be found in Section 5: Results from core models. Results from the logistic models can be found in Section 6: Results from logistic models.

4.1  The core models#

In order to control for variation in wealth not directly associated with variation in health, a series of control variables were used. These were grouped in two linear regression models, referred to as core models one and two. The core models were constructed to be used as a basis from which the measures of health could be considered.

Core model one includes age (and its square root), income, geographic region, ethnicity, highest qualification achieved, housing tenure, deprivation, gender and the composition of the household. These variables were chosen because of their relative independence to each other.

Core model two builds on core model one and also includes variables for whether the respondent was born in New Zealand, the number of years of paid employment, smoking habits, benefit receipt and being a student. The core models have the following form:

For core model one: In(NW) = ƒ(H,Z₁)

For core model two: In(NW) = ƒ(H,Z₁,Z₂)

Where:

NW = Net wealth scores

H = The particular health variable being considered

Z₁ = The control variables included in core model one

Z₂ = The new control variables introduced in core model two

All control variables were taken from Wave 2. The variables included in core model two but not in core model one were initially excluded owing to the probable lack of independence between themselves and other control variables. For example, whether the respondent was born in New Zealand was expected to be correlated with ethnicity. Years of paid employment and whether or not the respondent was a student were expected to be correlated with age.

Where control variables could be expressed in different forms, the significance and clarity of the variable were used to determine which form was included in the models. For completeness, the variables specified above were included in the models even if they were not found to be significant at the 10% level. Full regression tables of the core models can be found in Appendix A.

After modelling net wealth as the dependent variable, total assets and total liabilities were modelled with the same set of control variables. The aim was to identify whether changes in wealth were driven more by changes in assets or changes in liabilities.

Statistics New Zealand requires that all output is censored. All output has been weighted and counts rounded to the nearest hundred. Weighted counts of less than 1,000 are not released. Percentages are calculated after censorship. Both core models were run estimating robust standard errors. Data access was restricted to the Statistics New Zealand Datalab, where analysis was conducted using Stata Version 9.

4.1.1  Interpreting the results

The variables used in this study, including the health variables, are correlated with one another. Isolating these variables gives a simplistic representation of the factors that contribute to net wealth. In particular, we ignore the range of possible interactions between the health variables and the control variables, and interactions of the health variables with each other.

Many of the results from comparing health and wealth are given in terms of marginal effects. In the following discussion the marginal effects are talked about as “effects” on net wealth. It is important to keep in mind that these models do not prove or establish a causal link between the explanatory variables and wealth.

All marginal effects have been calculated at the mean value of the regressors. This means they apply to a theoretical person of mean age, income and number of years of paid employment. For categorical variables this theoretical person matches the weighted sample proportions. For example, because 78.7% of the longitudinal population were born in New Zealand and 21.3% of the longitudinal population were born overseas, the theoretical person uses 78.7% of the coefficient for being born in New Zealand and 21.3% of the coefficient for being born overseas. The means and proportions of the control variables used to calculate the marginal effects were computed from the entire longitudinal population and can be found in Appendix A, Appendix Table 3.

The regression methods used are known as mean regressions. The value of the dependent variable estimated by the regression model when each variable is set at its regression-sample-mean value will be its sample mean. Natural logarithms were used to transform net wealth before analysis. This greatly reduced the effects of outliers on the mean, though the median remained unchanged. As the model estimates the log of net wealth, the model's mean estimate will be the mean of the log of net wealth which, when transformed back, will be below the mean value of net wealth.

There are two age terms in the model: age and the square root of age.[10] For interpretation and application of these models the square root of mean age has been used in the calculation of the marginal effects. The estimates therefore apply to someone of mean age.[11]

4.1.2  Predicted values

Comparison of the observed logarithms of net wealth and those estimated by the models shows the model tends to provide reasonably good estimates for the population. Table 4 gives the percentiles of the logarithm of net wealth and the percentiles of the predicted logarithm of net wealth from core model two.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

The predicted values, for the logarithm of net wealth, for longitudinal respondents in the 95th and the 5th wealth percentiles tended to differ more from their actual values than for any other percentile of the population.

Figure 2 – Comparison of actual to predicted net wealth – core model two

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Figure 2 graphs the percentiles of net wealth, without logarithms, from Table 4. It should be noted than when transforming from the predicted logarithm of net wealth to predicted net wealth what seem like minor differences between the logarithms become much more significant differences between the transformed values. This is owing to the shape of the exponential curve. It would be prudent to limit the application of these models to levels of net wealth below $400,000.

4.2  The logistic models#

The use of logarithms in the core models excludes from analysis those respondents with negative or zero scores for the dependent variable as the logarithm of non-positive numbers is not defined. Logistic models consider the likelihood of an individual having non-positive net wealth and of having no liabilities. These individuals would have been excluded from the core model regressions of net wealth and total liabilities respectively.

Logistic models were chosen rather than probit models. This choice has little to no significance on the resulting fitted models. Logistic models are easier to calculate percentage point effects for and permit a larger variance than probit models.

4.2.1  Non-positive net wealth#

Non-positive net wealth arises from having total liabilities equal to or greater than total assets. After adjustments for student loan, 2.5% of the longitudinal population had non-positive net wealth.

This logistic model was built using the same descriptors as core model two with several changes: dwelling tenure and whether the respondent was a student were excluded from the model.[12] It was decided to exclude dwelling tenure as property is typically the largest contributor to net wealth so including it could be misleading. Whether the respondent was a student or not was excluded as adjustments had already been made for the presence of a student loan. The logistic model has the following form:

N_i = ƒ(H,Z'₁,Z'₂) for (i = 1, …, n)

Where:

N_i = 1 if the i^th respondent has negative or zero net wealth
    = 0 if the i^th respondent has net wealth greater than zero

H =The particular health variable being considered

Z'_k = The control variables from the core models

P_i = The probability of N_i = 1

The logistic model has shape:

4.2.2  Zero liabilities#

In order to consider total liabilities in the core models, the 28.2% of the longitudinal population who have zero reported liabilities had to be excluded. This makes the regression of total liabilities conditional on having liabilities. In order to provide a fuller picture we use a logistic model for whether an individual has liabilities or not.

The same logistic model as above was used with dwelling tenure and whether the respondent was a student or not being excluded. Dwelling tenure was excluded as the presence of a mortgage would have dominated the model. This logistic model has identical form to the above model specification with one change:

N_i = 1 if the i^th respondent has liabilities
    = 0 if the i^th respondent has zero liabilities

Property and mortgages are the most significant elements of assets and liabilities. According to SoFIE, of those who own property, three-quarters have more than 48% of their assets in property with a value of at least $70,500. Of those with a mortgage, for three-quarters of individuals their mortgage makes up at least 92% of their liabilities, with a value of at least $30,000.

Initial attempts were made to include dwelling tenure in both logistic models. However, as expected, it dominated the model with coefficients at least twice the magnitude of other control variables.

4.3  Modelling health#

This section provides some additional details that are supplementary to the regression models. These details inform parts of the results from the core models.

4.3.1  Health failures and the health survey models#

Wealth was measured at Wave 2 whereas measures of physical discomfort and psychological distress were recorded in Wave 3. This may be problematic if enough individuals experienced a substantial change in health between Waves 2 and 3. Deteriorating health between these waves could push some respondents from moderate (in Wave 2) to high discomfort and distress (in Wave 3). Alternatively, an improvement in health could lead to some individuals experiencing a reduction in their levels of discomfort and distress between Waves 2 and 3.

SoFIE contains a variable that may enable some of those in the former group to be identified.[13] Respondents were asked in Wave 3 whether they experienced an injury or illness, which restricted their usual activities, lasting seven days or more in the 12 months preceding the interview. Those who answered “yes” to this question were assumed to have experienced a health “failure”. For these respondents, their Wave 3 health is less likely to be a suitable proxy for their unobserved health in Wave 2.

This was incorporated by decomposing the measures of physical discomfort and psychological distress by the presence of a health failure. In the core models the health variable being considered changed from:

H = {physical discomfort, psychological distress}

to:

H = {physical discomfort by failure, psychological discomfort by failure}

4.3.2  Independence of the chronic conditions#

Independence of the chronic conditions was tested before including multiple conditions together in the core models. This was done by treating the occurrence of each chronic condition as a Bernoulli event. Seven of these conditions (excluding depression and schizophrenia) were combined to form a single multinomial model. The model has the following form:

The proportion of the longitudinal population with each condition was used as the best estimator for the probability of having the condition:

 

 

(c_i = 1) = proportion of the longitudinal population with c_i = 1

 

where c_i = 1 if the respondent has the i^th chronic condition.

From this, the probability of being diagnosed with every number and combination of conditions was calculated:

P(C = j) = Σ

 

(c₁ = 1) ·

 

(c₂ = 1) · … ·

 

(c₇ = 1)

 

with: Σc_i = j

where: C = the total number of chronic conditions for a respondent

These were grouped by the number of conditions in the diagnosis and multiplied by the weighted population total to give estimates for the entire longitudinal population.

Table 5 shows the results of this model and the actual observed results.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Inspection of the results shows the theoretical model to be a poor fit; this was confirmed by a goodness-of-fit test. The model overestimates the proportion of the longitudinal population with one or two conditions and underestimates the proportion of the longitudinal population with zero or three to seven conditions.

The failure of this model highlights the lack of independence among the conditions. This was not investigated further. A successful model would be expected to include age because older respondents are seen or seem to have a higher likelihood of developing all conditions.[14]

The lack of independence between chronic conditions means that including multiple conditions in the analysis tends to make one or more of the conditions redundant. Each core model was run once for each chronic condition, so as to avoid this. A summary variable that indicated how many chronic conditions each respondent suffered from was considered, but was discarded owing to poor fit.

4.3.3  Receipt of a health tested benefit#

SoFIE respondents were asked about the amount and sources of income they received over the last 12 months, including all forms of benefits. These benefits were separated into health and non-health benefits based on the requirements to qualify for each benefit.[15] SoFIE estimates 10% of the longitudinal population receive some form of health tested benefit.

An attempt was made to include the receipt of a health tested benefit in the core models. This resulted in many of the health descriptors becoming no longer significant at the 10% level and the coefficients of those variables that were still significant becoming significantly lower.

Much of the change caused by the inclusion of a health tested benefit will be due to colinearity. Because the receipt of a health tested benefit is dependent on a large number of factors including health, income and wealth, it is not a truly independent variable. Results from its inclusion can be found in Appendix C, Appendix Table 19.

5  Results from core models#

The core models provide a convenient base to analyse the association between health and wealth. As the regressions were run using the natural logarithm of wealth as the dependent variable, interpreting the coefficients is not straight-forward.

Positive coefficients imply the descriptor has a beneficial effect on wealth; negative coefficients imply a detrimental effect. Owing to the potential difficulty of interpreting the coefficients, marginal effects have been calculated. Full regression output (coefficients, standard errors and confidence intervals) can be found in the appendices.

To give context to the results that follow it is useful to have an idea of the marginal effects of the control variables from the core models. Ageing from about 42 to 47 is associated with an increase in wealth of $19,420 in core model one and $15,250 in core model two. In core model one, being of Māori or of Pacific Island descent is associated with having $31,670 and $62,490 less net wealth respectively than an individual of European descent. The difference between males and females is statistically significant, with males reporting $7,230 to $9,200 less net wealth than females. Receiving a non-health benefit is associated with having $12,960 less net wealth than those who do not receive any income from a government benefit. See Appendix A, Appendix Table 3 for a complete list of the marginal effects.

Certain control variables were frequently redundant in the core models. Geographic region, having children over five years and having previously been a smoker were frequently not significant at the 1% level.

5.1  The health surveys#

Physical functioning (PF) and the Kessler 10-item scale (K10) were used in core models one and two. Over both models, progressively worse health was linked to lower wealth for both physical and mental wellbeing.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=coefficients are significant at the 10% significance level. **=coefficients are significant at the 5% significance level. ***=coefficients are significant at the 1% significance level.
2. Marginal effects are calculated as the increase in net wealth for a change in the variable from the control with all other variables held at their mean.

Based on the marginal effects shown in Table 6 it is apparent that physical and mental wellbeing have differing effects. Those with moderate or high physical discomfort (9% and 6.5% of the longitudinal population respectively) experience similar levels of lower net wealth. Moderate psychological distress (14.5% of the longitudinal population) has a lower effect on net wealth, while high psychological distress (6.4%) has a larger effect. The coefficients for physical discomfort are not significantly different at the 10% level, whereas the coefficients for mental distress are significantly different at the 1% level.

Figure 3 shows the variation in the estimated level of net wealth for different levels of physical and mental wellbeing, with all other variables held at their mean. Net wealth is lower in every case where health status is worse.

Figure 3 – Levels of estimated net wealth by physical discomfort and psychological distress

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Levels of net wealth have been estimated for an individual with mean characteristics.
2. The means given here are forecasted means from the core models.
3. The differences between the bars for low and moderate, and for low and high discomfort and distress are identical to the marginal effects given in Table 6.

5.1.1  Health failures and the health surveys#

A respondent's health in Wave 3 may not be an accurate proxy for their health in Wave 2. The regressions involving the SF36 and K10 health measures were repeated, with the measures for physical discomfort and psychological distress decomposed by whether the respondent suffered a health failure or not.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. *=coefficients are significant at the 10% significance level. **=coefficients are significant at the 5% significance level. ***=coefficients are significant at the 1% significance level.

Table 7 gives the results after the decomposition. In all cases the impact on net wealth was greater for those recording a health failure. However, coefficient testing showed that only the coefficients for high physical discomfort, with and without health failure, were significantly different, and these only at the 10% level in core model two.

Further investigation revealed that of those who rated their own health fair or poor in Wave 2, over half of them experienced a health failure in the following 12 months, compared with less than a quarter of those who rated their health excellent or very good. This suggests that those who already had worse health were more likely to suffer from injuries in the subsequent year. Wealth may affect the speed at which people recover from injury or illness, so those with greater wealth may have been less likely to be recorded as having experienced health failures, which require the respondent to have an injury or illness lasting seven days or more.

Compared with those not reporting a health failure, those who report a health failure were more likely to report lower self-rated health in Wave 3 than in Wave 2. This holds across all categories of self-rated health. For example, of those with excellent self-rated health in Wave 2, 49% of those who suffered a health failure rated their health worse in Wave 3, while only 34% of those who did not suffer a health failure rated their health worse in Wave 3.[16]

5.2  Self-rated health#

In each wave, respondents rated their own health. The means and medians of net wealth for each of the self-rated health categories are given in Table 8. In all categories the raw mean exceeds the median by more than $50,000. This is owing to the skewed distribution of wealth.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. The modified mean is calculated as the mean of the logarithms of net wealth and then transformed back (raised to the exponential power). This reduces the effect of outliers.

While there is no clear overall systematic relationship between net wealth and self-rated health, Table 8 suggests that those with poor health do in fact have lower net wealth. However, these results do not control for the possible effect of other variables, in particular age. The lower mean and median for those reporting excellent health will be influenced by the number of young people with low wealth.

5.2.1  Results from self-rated health#

Self-rated health from Wave 2 was included in core models one and two. The results given in Table 9 show a progressive relationship with worse self-rated health being associated with lower net wealth. The marginal effects suggest that the relationship is stronger than was implied by the raw modified means; the difference between the raw modified means for very good and fair health is $1,800 and the difference between the marginal effects of very good and fair health is $21,950. Similarly, the difference between the raw modified means for excellent and poor health is $17,850 and the marginal effect of poor health is $46,150 (recall excellent health is the control).

Figure 4 shows net wealth declines systematically with worse self-rated health. For each core model the forecasted net wealth for an individual with the mean characteristics from the longitudinal population is given. In core model one, net wealth decreases from about $107,000 to just under $54,000 as health worsens from excellent to poor. In core model two, net wealth decreases from about $96,000 to just under $50,000.

The error bars provided in black give the ranges of forecasted net wealth for each health category if the corresponding coefficient is permitted to vary within its 95% confidence interval. The self-rated health coefficients are estimated with respect to excellent self-rated health, the error bars relate to the difference between excellent health and the corresponding category. They are not confidence intervals for the level of wealth by self-rated health.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=coefficients are significant at the 10% significance level. **=coefficients are significant at the 5% significance level. ***=coefficients are significant at the 1% significance level.
2. Marginal effects ($) are calculated as the increase in net wealth for a change in the variable from the control with all other variables held at their mean.
3. Coefficient testing shows all coefficients are distinct at the 5% level. The coefficient for very good health is distinct from zero (zero is equivalent to excellent health, the control) at the 1% level.

Figure 4 – Levels of estimated net wealth by self-rated health

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. The error bars (given in black) are calculated by estimating the log of net wealth using the bounds on the 95% confidence interval for each coefficient. This estimate is then transformed to net wealth.
2. Despite the overlap between some of the error bars, the coefficients are distinct at the 5% level. Any apparent overlap will be a result of changing from a logarithmic scale during the estimation process.
3. The means given here are those predicted by the core models for a respondent with average characteristics. They differ from those means given in Table 9 as the latter are raw means rather than predicted means.

5.3  The chronic conditions#

The chronic conditions were included as a series of binary indicators in core models one and two. Owing to the lack of independence between chronic conditions, these models were run separately for each chronic condition.

5.3.1  Results from the chronic conditions#

All chronic conditions had negative coefficients with only asthma and high blood pressure not significant at the 10% level.

Figure 5 shows the magnitude of the marginal effects for each chronic condition and For all conditions except asthma the marginal effect in core model two is less than the marginal effect in core model one. Some of this difference will be owing to the inclusion of additional control variables in core model two that are not in core model one. For example, smoking or receiving a non-health tested benefit may be associated with certain chronic conditions.

When interpreting these results it is important to note that they are derived by considering each chronic condition in turn, rather than simultaneously. The results can be directly compared to not having the chronic condition; comparison to other conditions can only be done indirectly. There were too few respondents with multiple chronic conditions to consider having a combination of conditions. The proportion of the longitudinal population with each chronic condition can be found in Appendix A, Appendix Table 3.

Figure 6 gives the prevalence weighted magnitude of the statistically significant chronic conditions.

Figure 5 – The estimated negative marginal effects of chronic conditions on wealth

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

For all conditions except asthma the marginal effect in core model two is less than the marginal effect in core model one. Some of this difference will be owing to the inclusion of additional control variables in core model two that are not in core model one. For example, smoking or receiving a non-health tested benefit may be associated with certain chronic conditions.

When interpreting these results it is important to note that they are derived by considering each chronic condition in turn, rather than simultaneously. The results can be directly compared to not having the chronic condition; comparison to other conditions can only be done indirectly. There were too few respondents with multiple chronic conditions to consider having a combination of conditions. The proportion of the longitudinal population with each chronic condition can be found in Appendix A, Appendix Table 3.

Figure 6 – The estimated negative marginal effects of chronic conditions on wealth – weighted by prevalence

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Results from the decomposition of the chronic conditions

All chronic conditions, other than depression and schizophrenia, were decomposed by whether diagnosis occurred within the last five years or more than five years ago. This was done to try to identify the immediate and long-term wealth effects associated with these chronic conditions. The presence of depression or schizophrenia could not be decomposed.

Figure 7 – The estimated negative marginal effects of chronic conditions on wealth – by timing of diagnosis – core model one

 

Source:  SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

It was unclear what effect this decomposition would have. Those who were diagnosed more than five years ago might no longer suffer the effects of their condition or might have experienced a greater effect on their wealth from having the condition for a longer period of time. Those with recent diagnoses were more likely to currently suffer from the effects of their condition, but any effect on their wealth might not yet be noticeable. Figure 7 illustrates the marginal effects on net wealth of each chronic condition depending on the timing of diagnosis.

Other than for asthma, the difference between the coefficients was not significant at the 10% level. Given the size of the differences between the marginal effects shown in Figure 7 we might have expected the differences for stroke and migraines to be significant. While the results are not generally statistically significant their relative magnitudes are consistent with the pattern that would be expected. Specifically, stroke produces permanent neurological deficit and would be expected to show a stronger effect on wealth when diagnosed more than five years previously. In contrast, asthma and migraine typically produce only short-lasting “health failures” and show a stronger effect when diagnosed within the last five years[17].

Having been diagnosed with asthma does not appear to have a long term association with a person’s wealth. This may be owing to respondents being diagnosed with asthma in their youth and growing out of it as they age.

The marginal effects for heart disease and diabetes are significant and similar regardless of the length of time since diagnosis.[18] This might suggest that these conditions have both an immediate and sustained impact on wealth.

5.4  Assets and liabilities#

Net wealth was decomposed into total assets and total liabilities. The aim was to determine how much of the change in net wealth was owing to changes in total assets or changes in total liabilities. The regression results for total assets follow a very similar form to net wealth, with all significant coefficients having the same sign and similar relative magnitude. Total liabilities were not as well explained by the core models as net wealth and total assets.[19] Only the results from core model one are considered here.

The marginal effects on net wealth are not equal to the difference between the marginal effects on assets and liabilities. This occurs as the use of logarithms to transform the dependent variable results in different sample populations for each regression. To obtain such a model would require using the same sample population or solving all three regressions at once using simultaneous equations.

It is common to observe a rise in assets with a rise in liabilities or a fall in assets with a fall in liabilities associated with the same descriptor. Only age, living in a dwelling owned without a mortgage and having children under five report assets and liabilities moving in opposite directions where both coefficients are significant.[20]

5.4.1  Assets, liabilities and health descriptors#

Table 10 gives the marginal effects of health on wealth, assets and liabilities. Marginal effects are used as they are more comparable between regressions than coefficients. However, the comparison is still flawed as the marginal effects are calculated from regressions with different samples.

Lower net wealth owing to ill health appears to be associated with lower total assets. Most of the coefficients for the amount of debt held are not significant at the 10% level, and the magnitude of the marginal effects for total liabilities with significant coefficients is often less than one-tenth the size of the corresponding marginal effects for total assets. The logistic model for having liabilities, in Section 6.2, suggests only certain measures of ill health are associated with whether people are in debt or not.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=coefficients are significant at the 10% significance level. **=coefficients are significant at the 5% significance level. ***=coefficients are significant at the 1% significance level.
2. Marginal effects ($) are calculated as the increase in the dependent variable for a change in the variable with all other variables held at their mean. For categorical variables this is a change from the control.

The liability model is conditional on having non-zero liabilities. These results should be considered with the logistic model for whether people have or do not have liabilities.

6  Results from logistic models#

The use of logarithms to transform the dependent variable in the above analysis excluded all respondents with zero or negative scores. To provide a fuller picture and to consider those respondents with zero or negative responses, a pair of logistic models was constructed.

6.1  The likelihood of negative or zero net wealth#

This logistic model gives an idea of the characteristics associated with those individuals whose liabilities exceed their assets. These results should be considered in conjunction with the results for net wealth.

After the regression the only control variables significant at the 10% level were age, geographic region, ethnicity and being a current smoker. The mean longitudinal respondent has a 2.6% chance of having non-positive net wealth. In absolute terms few coefficients had a percentage point effect of 2% or greater. However, in relative terms, a percentage point effect of 2.0% is significant as the likelihood of having non-positive net wealth close to doubles. Within the sensible bounds of the model, no individual in the longitudinal population can be estimated to have more than a 22% likelihood of having negative or zero net wealth.

To provide some context for the results below it is useful to know that being a current smoker results in a percentage point increase of 1.1 to the probability of having non-positive net wealth; being of Māori or Pacific Island descent results in percentage point increases of 1.6 and 2.4 respectively; ageing from 42 to 47 results in a 0.7 percentage point decrease. The logistic model, before the inclusion of health descriptors, can be found in Appendix C, Appendix Table 20.

The health descriptors were included in this logistic model in the same form as they were included in the core models. Table 11 gives the coefficients of the health survey descriptors and their percentage point effects. Percentage point effects for the health variables should be used for comparison as the coefficients are not directly comparable.

As an illustration of how to interpret the percentage point effects in Table 11, consider the effect of an increase in psychological distress on the probability of having non-positive net wealth: the probability of a person with average characteristics without any psychological distress of having non-positive net wealth is 2.32%. The probability of the same person but with high levels of psychological distress is 4.48%. The difference between these two (4.48 - 2.33) is 2.15; this is the percentage point difference shown in the table.

As in the core models the health surveys were decomposed by whether or not the respondent had experienced a health failure in the previous 12 months. There was no significant difference between the coefficients before and after the decomposition. For self-rated health, none of the adjacent categories are significantly different from each other at the 5% level.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=significant at the 10% significance level. **=significant at the 5% significance level. ***=significant at the 1% significance level.
2. Percentage point changes are calculated as the increase in percentage points of the probability to have non-positive net wealth for a change in the variable with all other variables held at their mean. For categorical variables this is a change from the control.

Many health variables were significant and resulted in at least a 1 percentage point increase on the likelihood of non-positive net wealth.[21] The health variables that were not significant in the logistic regression were those that were either not significant or had smaller marginal effects in the core models. The exceptions to this are diabetes and stroke, which have the greatest standard error of the chronic conditions.

The results from the core and logistic regressions show that worse health is associated with lower wealth and a higher likelihood of negative or zero net wealth. This suggests that the costs of ill health can exceed an individual's ability to meet them and could potentially draw them into undesirable debt.

6.2  The likelihood of having liabilities#

This logistic model should be considered in conjunction with the regression of the logarithm of total liabilities. It is expected that we will see most variables to have the same direction in both models (the level of total liabilities held should move with the likelihood of holding liabilities).

An individual's geographic region, having children over the age of five and receiving a benefit were the only descriptors that were not significant at the 10% level. A member of the longitudinal population with mean characteristics has an 83.4% chance of having liabilities. The likelihood of having liabilities increases until people are about 40 years of age and decreases from 40 onwards. By the age of 80 there is just more than a 15% chance of having liabilities at the mean.

To provide some context for the results below it is useful to know that having a university qualification increases the likelihood of having liabilities by 6.7 percentage points. Being moderately to heavily deprived (median to worst categories) are associated with a 3.7 to 4.4 percentage point increase in the likelihood of having liabilities. The logistic model, before the inclusion of health descriptors, can be found in Appendix C, Appendix Table 22.

The health descriptors were included in this logistic model in the same form as they are found in the core models.

Table 12 gives the coefficients and percentage point effects from these regressions. Percentage point effects for the health variables should be used for comparison as the coefficients are not directly comparable.

All significant health variables have positive coefficients. Only high physical discomfort and the chronic conditions (excluding stroke) appear to have a significant effect on whether a person has liabilities or not and all are associated with an increase in the likelihood of having liabilities.

Of the health descriptors, migraines have the greatest percentage point effect of 3.44. This effect is smaller than those associated with ethnicity, education, deprivation and family composition and is relatively minor compared to the 83.4% probability of having liabilities at the mean. Ageing from 25 to 27 has the same percentage point increase on the likelihood of having liabilities as migraines. Ageing from 56 to 58 has the reverse percentage point effect of being diagnosed with migraines.

The weak association between health and whether a member of the longitudinal population is in debt may be owing to when they enter into debt and when their health begins to deteriorate. Individuals with ill health may be less inclined to enter into debt, while healthy individuals may enter into debt and later find their health deteriorating. The logistic model will be unable to detect this as it is based on cross-section data.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=significant at the 10% significance level. **=significant at the 5% significance level. ***=significant at the 1% significance level.
2. Percentage point changes are calculated as the increase in percentage points of the probability to have zero liabilities for a change in the variable with all other variables held at their mean. For categorical variables this is a change from the control.

7  Conclusions#

This paper has examined the relationship between net wealth and health. Net wealth was modelled as a function of health with the inclusion of a range of control variables. Across a range of health measures it found a fairly clear association, with net wealth tending to be lower for individuals with poorer health.

Results using separate measures of physical and mental wellbeing showed both to be associated with lower net wealth. An individual with moderate physical discomfort was estimated to have $11,330 less wealth than an individual with only low physical discomfort, holding other factors constant at their mean values. This difference was larger for an individual with high physical discomfort who was estimated to have $12,890 less wealth. For mental health, a mean individual with moderate distress had an estimated $5,330 less net wealth than the same individual with low distress. The difference between high and low distress was $28,850 less net wealth.

The measures of physical and mental wellbeing were decomposed by whether the respondent had experienced a short-term health failure, defined as responding “yes” to having an illness or injury, which hindered their normal activities, lasting at least seven days in the previous 12 months. The results showed that the marginal effect of ill health on wealth for those who suffered a health failure was more negative than for those who did not suffer a health failure. Further investigation revealed that those who did not suffer from health failures had, on average, better self-rated health and greater wealth.

Respondent self-rated health was significant in almost all analyses. Self-rated health was used in order to provide a wider perspective of a person's wellbeing than targeted questioning alone. Respondents were asked to rate their health as excellent, very good, good, fair or poor. The worse a respondent rated their health, the progressively lower their wealth. A person with average characteristics who rated their health very good was estimated to have $11,400 less net wealth than a person with the same characteristics who rated their health excellent. The difference between good and excellent health was $25,760, between fair and excellent health was $35,500 and between poor and excellent health was $53,000.

Eight chronic conditions were considered: asthma, high blood pressure, high cholesterol, heart disease, diabetes, stroke, migraines and the presence of either depression or schizophrenia. Of these, heart disease, diabetes, stroke and the presence of either depression or schizophrenia had the greatest marginal effects on wealth[22]. Those with these conditions had $10,930 to $19,150 lower net wealth, at the mean, than those who did not. Owing to a lack of independence between the conditions, and the small number of respondents with multiple conditions, the effects were all measured separately.

The marginal effects of the six statistically significant chronic conditions were weighted by prevalence of the condition in the longitudinal population. The magnitude of the marginal effects for heart disease and diabetes were reduced to similar levels of effect, per head of population, as high cholesterol and migraines, about $600 per person. The presence of depression or schizophrenia had twice the effect per head of population with a predicted cost of $1,380.

The chronic conditions were decomposed by whether the respondent had been diagnosed within the last five years, or more than five years ago. The variable for depression or schizophrenia was not able to be decomposed in this way. Asthma has a statistically greater effect on net wealth in the first five years. Diabetes and heart disease were the only conditions that were significant regardless of how long since diagnosis.

Net wealth was decomposed into assets and liabilities. Models were estimated for assets and liabilities separately using the same control variables as were used in the regression models for net wealth. From this, the majority of the lower net wealth associated with ill health appears to be the result of lower total assets with total liabilities remaining reasonably constant.

A logistic model was estimated for the probability of having negative net wealth. The likelihood of having negative net wealth was found to increase as self-rated health decreased, and to be higher for those with high psychological distress compared to those with lower psychological distress. Four of the chronic conditions were found to have significant association with a higher likelihood of negative net wealth: high cholesterol, heart disease, migraines and depression or schizophrenia. No relationship with physical health was found.

A logistic model for the probability of having liabilities was also estimated. It was found that mental wellbeing and the levels of self-rated health were largely not significant in determining whether a respondent has liabilities or not. But high physical discomfort and being diagnosed with a chronic condition, other than stroke, was found to increase the likelihood of having liabilities. Of these, migraines had the greatest percentage point effect. However this was dwarfed by the control variables, which frequently had more than twice the percentage point effect.

This analysis is only able to detect the association between health and wealth; identifying the direction of causality is not possible given the techniques used. There is the potential for health to affect wealth and for wealth to affect health. It is also difficult to be certain that the relationship detected is not owing to unobserved variables having an effect on both health and wealth. Untangling the complex relationship between health and wealth remains an area for future research. The longitudinal nature of SoFIE will eventually enable more complex modelling techniques to be applied to the data.

Despite not resolving the question of causality, this paper does provide evidence that those with poor health tend to be financially worse off. Those with ill health may have lower wealth owing to a variety of reasons, including the cost of treatment, reduced earnings and higher cost of living. How much of the cost to treat ill health is borne by those who suffer from ill health will influence the apparent magnitude of this relationship.

As with any study of this nature, the robustness of the results reflects in part the quality of the data and inevitably there will be errors of measurement in the data. The reporting of assets and liabilities relies on recall by the respondents, and some items may have been overlooked. Furthermore, it is recognised that very high net wealth individuals may not be proportionately represented in SoFIE. While self-reported health status has been widely used, there exists the potential for bias in reporting if the respondents have a perception that in responding to an official survey their present benefits might be in jeopardy.

A potential by-product from public investment in improving the health of an individual could be higher levels of net wealth. Typically those with better health status tend to have greater productivity, higher incomes and longer working lives, all of which provide an opportunity to accumulate greater net wealth. However, the impact of the taxes needed to fund such investment and the labour force participation rates of those targeted would need to be considered before affirming that such investment would have net positive benefits.

An implication of having benefits that are asset tested against net wealth is that those with poor health will have a tendency to be overrepresented among recipients. So it is expected that a greater proportion of those receiving a benefit will have ill health, compared to the proportion of the total population with ill health.

References#

Andrews, Gavin and Tim Slade (2001) "Interpreting scores on the Kessler psychological distress scale (K10)." Australian and New Zealand Journal of Public Health 25(6): 494-497.

Bryant, John, Audrey Teasdale, Martin Tobias, Jit Cheung and Mhairi McHugh (2004) Population ageing and government health expenditures in New Zealand, 1951-2051. Wellington: The Treasury.

Carter, Kristie, Michael Hayward and Ken Richardson (2008) "SoFIE-health baseline report: Study design and associations of social factors and health in waves 1 to 3. SoFIE-health Report 2." Wellington, University of Otago.

Carter, K.N., T.Blakely, S.Collings, F.I.Gunasekara and K. Richardson (2009a) "What is the association between wealth and mental health?" J Epidemiology and Community Health 63(3): 221-226.

Carter, K.N., M.Cronin, T.Blakely, M.Hayward and K.Richardson (2009b) "Cohort profile: Survey of Families, Income and Employment (SoFIE) and Health Extension (SoFIE-health). International Journal of Epidemiology. Published online May 28, :dyp215.

Case, Anne (2001) "Does money protect health status? Evidence from South African pensions." Working Paper 8495, National Bureau of Economic Research.

Case, Anne and Angus Deaton (2003) "Broken down by work and sex: How our health declines." Working Paper 9821, National Bureau of Economic Research.

Creedy, John and Li Tan (2007) "The relationship between personal income and net worth in Australia." Australian Economic Review 40(2): 165-181.

Diener, Ed, Eunkook M. Suh, Richard E. Lucas and Heidi L. Smith (1999) "Subjective well-being: Three decades of progress." Psychological Bulletin 125: 276-302.

Headey, Bruce and A.J. Wearing (1992) Understanding happiness: A theory of subjective well-being, Melbourne: Longman Cheshire.

Headey, Bruce, Gary Marks and Mark Wooden (2005) "The structure and distribution of household wealth in Australia." Australian Economic Review 38(2): 159-175.

Headey, Bruce and Mark Wooden (2004) "The effects of wealth and income on subjective well-being and ill-being." Economic Record 80: S24-33.

Henderson, Katherine and Grant M Scobie (2009) "Household debt in New Zealand." Working Paper 09/03, Wellington: The Treasury.

Holt, Heather (2010) " Health and labour force participation"."Working Paper 10/03. Wellington, New Zealand: The Treasury.

Meer, Jonathan, Douglas Miller and Harvey Rosen (2003) "Exploring the health-wealth nexus." Working Paper 9554, National Bureau of Economic Research.

Phongsavan, Philayrath, Tien Chey, Adrian Bauman and Robert Brooks (2006) "Social capital, socio-economic status and psychological distress among Australian adults." Social Science & Medicine 63: 2546-2561.

Redhead, K. A. and L.C. Rose (1999) "Reasons for problem debt: Theory and evidence." Working Paper No. 99.12 (Palmerston North: Massey University).

Scobie, Grant, John K. Gibson and Trinh V. T. Le (2005) Household wealth in New Zealand. (Wellington: Institute of Policy Studies)

Scott, M. Kate, Diana Sarfati, Martin I. Tobias and Stephen J. Haslett (1999) "Sf-36 health survey reliability, validity and norms for New Zealand." Australian and New Zealand Journal of Public Health 23: 401-406.

Scott, M. Kate, Diana Sarfati, Martin I. Tobias and Stephen J. Haslett (2000) "A challenge to the cross-cultural validity of the SF-36 health survey: Factor structure in Maori, Pacific and New Zealand European ethnic groups." Social Science & Medicine 51(11): 1655-1664.

Wu, Stephen and Harvey Rosen (2003) "Portfolio choice and health status." Working Paper 9453, National Bureau of Economic Research.

Appendix A#

Appendix A contains the results from the regressions of the core models before health variables were introduced.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix A (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Where units are provided the measurement is a mean. Where no units are provided the value is the proportion of the longitudinal population with the characteristic. The proportion of the longitudinal population with the control characteristic has also been provided for all descriptors other than the chronic conditions.
2. Proportions may not sum to 1 owing to rounding after calculation. All proportions are rounded to four decimal places.
3. For the calculation of marginal effects of the chronic conditions, the regression coefficients for the non-chronic descriptors were taken from the regression of asthma. This ensures all the marginal effects have the same control point.

Appendix B#

Appendix B contains the results from the regressions including health variables. Only the health variables are reported here. The control variables differ from their values in the core models but these differences are minor and are not the focus of this paper, hence they are omitted.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix B (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix B (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix B (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Each chronic condition was regressed separately. All eight are shown together here for compactness.
2. Regression characteristics provided give the minimum and maximum for each regression characteristic over the eight regressions.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Each chronic condition was regressed separately. All eight are shown together here for compactness.
2. Regression characteristics provided give the minimum and maximum for each regression characteristic over the eight regressions.

Appendix B (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Each chronic condition was regressed separately. All seven are shown together here for compactness. Depression or schizophrenia was not able to be decomposed.
2. Regression characteristics provided give the minimum and maximum for each regression characteristic over the seven regressions.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. Each chronic condition was regressed separately. All seven are shown together here for compactness. Depression or schizophrenia was not able to be decomposed.
2. Regression characteristics provided give the minimum and maximum for each regression characteristic over the seven regressions.

Appendix C#

Asset and liability decompositions#

This section contains the results from when the logarithm of total assets and the logarithm of total liabilities were run through core model one.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix C (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix C (continued)#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. *=coefficients are significant at the 10% significance level. **=coefficients are significant at the 5% significance level. ***=coefficients are significant at the 1% significance level.
2. Marginal effects ($) are calculated as the increase in the dependent variable for a change in the variable with all other variables held at their mean. For categorical variables this is a change from the control.
3. The liability model is conditional on having non-zero liabilities. These results should be considered with the logistic model for whether people have or do not have liabilities.

Inclusion of health tested benefit receipt

The following table gives the results from the health descriptors with the inclusion of a variable for the receipt of a health tested benefit.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Appendix C (continued)#

General logistic models#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. Percentage point changes are calculated as the increase in percentage points of the probability to have non-positive net wealth for a change in the variable with all other variables held at their mean. For categorical variables this is a change from the control. For age and years of paid employment this is a five-year change, for personal income a change of $5,000.

The probability of a European with average characteristics having non-positive net wealth is 2.48%. The probability of the same person if they were Māori is 4.08%. The difference between these two (4.08 - 2.48) is 1.60, the percentage point difference.

Source and Note: See Appendix Table 20.

Appendix C (continued)#

Source and Note: See Appendix Table 20.

The probability of a person with average characteristics who is European having liabilities is 84.71%. The probability of the same person if they were Māori is 81.63%. The difference between these two (81.63 - 84.71) is -3.08, the percentage point difference.

Source and Note: See Appendix Table 20.

Appendix C (continued)#

Changes in self-rated health decomposed by health failures#

This section contains supporting tables for Section 5.1.1.

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Notes:

1. The proportions given on the right hand side show the proportion of the longitudinal population whose self-rated health improved, worsened or stayed the same between waves.
2. S = results censored due to too few observations. These cells contained less than 1,000 weighted observations before release.

From the above tables it can be seen that a greater proportion of the longitudinal population who suffer a short-term health failure rate their health worse between waves, and a smaller proportion rate their health better between waves, than those who do not suffer a health failure.

Appendix C (continued)#

Health failures and self-rated health#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Of individuals with fair or poor self-rated health more than half suffered a health failure between Waves 2 and 3. Only 18% of those with excellent self-rated health suffered a health failure between Waves 2 and 3.

Health failures could be caused by injury or an illness lasting more than a week. Whether the health failure was caused by injury or illness was not considered in this analysis. Consideration of the difference between these two may be useful in further research.

Wealth and health failures#

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Note:

1. These measures of net wealth have not had student loans offset against them.

For all percentiles, those who experienced a health failure have lower net wealth than those who did not experience a health failure. The difference between the net wealth of those who experienced a health failure and those who did not increases as the overall level of net wealth increases. This suggests that health failures are associated with lower net wealth. However, these results do not control for the possible effect of other variables.

Appendix C (continued)#

Tables of correlations#

This section contains tables of correlations to support references to the correlation between certain variables throughout this paper.

The SF36 physical health measures are: physical functioning (PF), role limitations owing to physical functioning (RP), bodily pain index (BP), general health perceptions (GH) and the physical component summary measure (PCS).

The SF36 mental health measures are: social functioning (SF), role limitations owing to emotional functioning (RE), general mental health perceptions (MH), vitality (VT) and the mental component summary measure (MCS).

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand

Source: SoFIE Waves 1–3, OSMs, longitudinal weights, supplied by Statistics New Zealand