The extent to which financial behaviour can explain over-indebtedness amongst New Zealand families
4. The confounding effects of family and financial circumstances
There are a number of factors that could explain the likelihood of being over-indebted and therefore confound the impact financial behaviour has on over-indebtedness.
This section outlines a more realistic model of over-indebtedness, the theoretical and empirical basis for the additional variables, and the suitability of LSS 2004 data to approximate them.
4.1. Theory and evidence
This section also summarises the key findings in the Families Commission and Retirement Commissions’ 2008 report (Legge & Heynes, 2008).
The main finding was that it is difficult to isolate a single indicator of indebtedness or over-indebtedness. Of the studies reviewed, a number of empirical relationships were identified. It is important to note that these studies were based on different datasets (in time and place) as well as different definitions and types of debt (or problem debt). In some cases where empirical relationships were found, it was difficult to establish the extent to which confounding factors were controlled for. None of the studies are thus directly comparable and at best the study identified a number of hypotheses for testing with New Zealand data. Hence the current study.
Debt usage or participation in the debt market is strongly correlated with age in a number of overseas studies (Tudela & Young, 2004; Balmer, Pleasance, Buck & Walker, 2005; Kempson, 2002). That is, people are increasingly likely to borrow over the first half of their working life when they have fewer resources and more demands on those resources (due to the costs of creating and raising a family), and are decreasingly likely to borrow over the second half of their working life. The relationship between the amount of debt people take on and age, however, has not been clearly established in the literature (Lunt & Livingstone, 1992; Del-Río & Young, 2005). This age effect is presumably confounded by income and wealth – older people may be less likely to use debt when they are older, but may be able to finance much higher levels of debt when they do.
The impact that family size or the number of children has on indebtedness and over-indebtedness is also inconclusive, most likely due to the confounding effects of income and wealth (Lunt & Livingstone, 1992; Lindqvist, 1981; Kempson, McKay & Willitts, 2004; Valins, 2004). However, there is some evidence of a positive, causal correlation with relationship breakdown and over-indebtedness (Kempson et al, 2004; Balmer et al, 2005).
As expected, income has been found to be a strong indicator of the amount of debt people take on and of problem debt, but a poor indicator of debt usage or participation in the debt market (Lunt & Livingstone, 1992).
Having an optimistic view of one’s future financial position and having higher qualifications has been associated with increased use of unsecured debt, but it is not clear whether the effects of income and borrowing for education have been adequately controlled for (Del-Río & Young, 2005).
Unemployment, benefits receipt and long-term illness or disability have all been found to be positively associated with over-indebtedness (Kempson, 2002; Balmer et al, 2005).
No overseas evidence has been found linking ethnicity with indebtedness and/or over-indebtedness. Age, income and wealth are likely to be significant confounding factors as Māori and Pacific peoples in New Zealand have young age demographic structures compared with the general population (SNZ, 2009) and are disproportionately represented in low-income (MSD, 2008) and net-worth statistics (Cheung, 2007). It is possible, however, that ethnicity might indirectly influence indebtedness and/or over-indebtedness to the extent that culture influences financial behaviour (providing financial support to extended family, for example). There is some evidence that this might be worth exploring, but examining what influences behaviour is beyond the scope of this current research.
4.2. Extended model
Based on this evidence, the basic model outlined in equation (1) can more realistically be re-specified as follows:
Prob(over-indebted) =
α + β1 financial behaviour
+ β2 family characteristics
+ β3 financial circumstances + ε. (2)
The slope coefficients (βs) represent the marginal or ‘pure’ (rather than potentially confounded) effect that each of these explanatory variables has on the likelihood of families being over-indebted. If financial behaviour does have a marginal effect on the likelihood of being over-indebted, the coefficient would be non-zero with a reasonable degree of significance. If the effect is not confounded by other factors such as family characteristics and financial circumstances, the coefficient β1 estimated in (2) should be similar to the coefficient β estimated in (1).
4.3. Confounding variables in the Living Standards Survey 2004
Family characteristics
Variables on age, relationship status, recent break-up and number of children have been constructed from the LSS 2004. It is these underlying family characteristics, rather than any specific family type (such as sole parenthood) that the literature suggests influence over-indebtedness.[9] These influences need to be removed in order to observe any pure effect financial behaviour has on over-indebtedness.
‘Age’ is a continuous variable and is the reported age of the respondent in the LSS 2004 dataset[10], although an age of the family could be derived from the dataset in future analysis. ‘Age2’ is also included in the regression to reflect a possible quadratic (or ‘lifecycle’) relationship with financial strain. This relationship would have a ‘U’ shape: young and old families are more likely to experience financial strain as respectively they are at the beginning and end of their working lives.
‘Relationship status’ is the reported social marital status of the respondent in the LSS 2004 dataset.[11] This is a dummy variable with values 0 for singles and 1 for couples. Couple families are expected to be less likely to experience financial strain than single families, primarily due to access to resources. Therefore a negative relationship is expected between this variable and financial strain.
‘Recent break-up’ represents respondents who reported experiencing a break-up of a marriage or de facto relationship.[12] This is also a dummy variable with values 0 for no and 1 for yes. As mentioned earlier, relationship break-up was found to be positively and causally linked with problem debt in the United Kingdom.
‘Number of children’ is the count of children in the respondent’s family group, reported as having any of the following relationships with the respondent: biological, step-child, adopted-biological, adopted-not biological, fostered, whangai.[13] Having lots of children is expected to put pressure on a family’s resources and is therefore expected to be positively related to financial strain. Legge & Heynes (2008) acknowledged that family age, income and wealth are likely to confound any relationship the number of children could have with financial strain, hence mixed international evidence. A positive but weak relationship is therefore expected with financial strain.
Financial circumstances
Qualification data arguably represent family characteristics and financial circumstances. After preliminary testing, two dummy variables have been created from the respondent’s reported highest qualification in the LSS 2004 dataset:[14] the first is ‘any qualification’ (0=none, 1=any); the second is any ‘post school qualification’ (0=none or school only, 1=higher than school). Both variables are expected to be negatively related to financial strain, but positively related to family income.
‘Family net income’ is derived from reported responses to the LSS 2004 question:[15] “what is your best estimate of (a) your personal total income (for the past 12 months), and (b) your partner’s total income (for the past 12 months)?” The responses were coded into income bands provided in the survey questionnaire, along with information on whether they refer to before or after tax amounts. These data were converted to midpoints (using an estimate for the upper band) and net tax figures were produced.[16] The result was a continuous variable based on midpoint data. The variable has also been divided by 1000 dollars to make the marginal effect more obvious. A strong negative association is expected with financial strain, as per the earlier discussion.
It is worth noting that the income distribution generated by this variable suggests that 50 percent of families have net incomes of less than $16,250. As expected, low-income families are also disproportionately represented by single families (with one income earner) and high-income families by couple families (with up to two income earners). It is possible that there is some under-reporting of income, and that income is systematically being under-reported by single families. Despite the apparent wording in the survey questionnaire to the contrary, it is plausible that low-income families are not fully reporting (or possibly accessing) social assistance entitlements or tax credits. Establishing whether this is or is not the case is outside the scope of this research, but does need to be considered as a potential data bias. The implication of such a bias is that the relationship between EFU income and financial strain could be over-estimated.[17]
A dummy variable is also derived for ‘benefit receipt’, which captures any benefit received by the respondent[18] or partner[19] in the last 12 months (0=none, 1=any). To some extent this should remove any possible bias caused by any systematic under-reporting of income by low-income families. A positive relationship with financial strain is expected with this variable.
Finally, ‘net debt’ is derived from the total reported debt of the family less the family’s total reported assets. Family assets comprise the latest government or rating valuation of property[20] and the total value of current savings and investments.[21] Family liabilities comprise the total outstanding and unpaid debt owed for hire-purchases, credit cards, overdue fines, overdue bills and overpaid benefits,[22] the amount currently owing on student loans,[23] and the total debt owed on accommodation.[24] According to the data, 30 percent of families have ‘net debt’.[25]
The result is a continuous variable based on midpoints, similar to that derived for family or EFU net income. As with income, this variable has been divided by 10,000 dollars to more easily interpret any marginal effect. A strong negative relationship is expected with financial strain. However, net debt could also represent illiquid resources, so it is conceivable that a family with low net debt could still experience financial strain in the short term.
4.4. Estimation
The extended model, based on the variables selected from the LSS 2004 dataset, can be re-specified as follows:
Ln(problem debt) =
α + β1 age + β2 age2 + β3 relationship status
+ β4 recent break-up + β5 number of children
+ β6 family net income + β7 net debt
+ β8 insured + β9 ‘spender not saver’ + ε. (2a)
As discussed with the basic model, finding a statistically significant relationship between any of the explanatory variables and problem debt does not automatically suggest a causal relationship. The theory and data provide the strongest basis for asserting a causal relationship. The explanatory variables in this model are mostly determined prior to or concurrently with the reported experience of financial strain, which had to be in the last 12 months, so a weak causal relationship is assumed with a significant result.
It should also be noted that the sample data used in the regressions are unweighted; and that ‘don’t know’ or ‘refused’ responses were omitted from the regression analysis. The regressions were therefore run with reduced sample sizes, varying according to which variables were included in the regressions.
As with the basic model, the extended model being estimated is effectively a logit model, as the dependent variable (problem debt) is a binary or dummy variable with values 0 and 1. In other words, the problem the models are trying to solve is the probability or likelihood of a family experiencing problem debt – ie of a family getting a value of 1.
4.5. Results
Figure 7 shows the results of three sets of multivariate regressions against the three different definitions of problem debt. The behavioural variable ‘spender not saver’ consistently emerges as having the strongest marginal effect. The marginal effect of this variable, ranges between 0.13 and 0.23. This means that being a ‘spender not saver’ (ie having low self-control and external locus of control) increases the likelihood of a family experiencing problem debt (however it is defined) by between 13 and 23 percent, controlling for other variables. This is a significant finding.
The presence of other variables only appears to halve (rather than remove) the marginal effect of the ‘spender not saver’ variable. Figure 6 indicates that regressing this variable on its own appears to account for up to 42 percent of the likelihood of experiencing financial strain.
The effect of insurance on the likelihood of experiencing problem debt, on the other hand, is significantly reduced when other explanatory factors are accounted for.
Note: each column represents the result of a separate multivariate regression
Statistical significance: *10%, **=5%, *** 1%
4.6. Summary
Even after controlling for a range of variables that might also explain problem debt, the behavioural variable ‘spender not saver’ emerges as having a strong independent effect, irrespective of the definition of problem debt.
Footnotes
- [9]
- This is a similar rationale for not including ethnicity in the analysis. See note 23. [Return to reference]
- [10]
- Q10_a1, LSS 2004 Questionnaire. [Return to reference]
- [11]
- Q4 a2–a10 LSS 2004 Questionnaire = ‘spouse or partner’, for adults in the respondent’s family group. [Return to reference]
- [12]
- Q69 LSS 2004 Questionnaire = ‘once’ or ‘more than once’ and reported that break up occurring less than a year ago (Q70). [Return to reference]
- [13]
- Q4 c1–c10, LSS 2004 Questionnaire. [Return to reference]
- [14]
- Q19, LSS 2004 Questionnaire. The highest family qualification has not been derived, but could be with future analysis. [Return to reference]
- [15]
- Q109, LSS 2004 Questionnaire. [Return to reference]
- [16]
- Tax rates (incl ACC levy): up to 38,000 = 19.5 (20.9); 38,001–60,000 = 33 (34.4); 60,001 = 39 (40.4). [Return to reference]
- [17]
- Another consideration with using EFU income, mentioned in Section 2.1, is that family income (and assets) may be a poor predictor of financial strain if families in financial strain live in multi-EFU households. One might expect an economic model to overestimate the relationship between income and financial strain if families in financial strain live in multi-EFU households, as the data will understate access to resources and consumption sharing (or economies of scale). On the other hand, ‘financial strain’ may in fact be capturing this resource-sharing between EFUs. [Return to reference]
- [18]
- Q110, LSS 2004 Questionnaire. [Return to reference]
- [19]
- Q111, LSS 2004 Questionnaire. [Return to reference]
- [20]
- Q127, LSS 2004 Questionnaire. [Return to reference]
- [21]
- Q137, LSS 2004 Questionnaire. [Return to reference]
- [22]
- Q151, LSS 2004 Questionnaire. [Return to reference]
- [23]
- Q141, LSS 2004 Questionnaire. [Return to reference]
- [24]
- Q130, LSS 2004 Questionnaire. Note that total debt figures do not capture overdrafts, due to the specific wording in the LSS 2004 questionnaire. See note 3, above. [Return to reference]
- [25]
- This proportion seems high, but may be due to the computation of EFU rather than household or individual net debt. In fact, this seems more plausible in relation to reporting of debt and assets than in relation to income. Young single families are arguably more likely to incur debt on the basis of anticipated future earnings (eg student loans) or an implicit/explicit guarantee provided by older family members, particularly in multi-EFU households. It should also be noted that the measure does not capture household items. [Return to reference]