Causation and income dynamics of people in disadvantage
The surveys above asked the question, “What causes poverty?” But before we, as researchers and policy makers, can answer this question, we must first understand how causal relationships “work.” Those familiar with the nature of causal relationships and the associated research methods may skip this section, as it serves as an introductory primer to the area.
Let’s start with an example. Poverty is linked with almost every negative outcome imaginable, both now and in the future. These span educational, health, economic, and social domains and include “brain and cognitive development, current and future mental and physical health, lifespan, child abuse and neglect, learning, future earnings, successful relationships, and so on.” The academic literature here is robust: there is a strong consensus that these relationships exist.
But just because poverty is related to these negative outcomes, does not necessarily mean that poverty is the primary cause driving them. Causation, that is, how and why two factors are related, is much more difficult to come by than correlation. It isn’t as simple as observing X occurring and watching Y result.
More generally, when two variables coincide, three possibilities arise:
- Causation is not involved at all. X doesn’t cause Y. Sometimes, while two variables are related and follow one another there is no causal relationship. Spring, for example, always follows Winter, but Winter does not cause or produce Spring in any meaningful way.
- The observed relationship is a result of outside factors: Both X and Y are caused by something else. Other times, something else causes the changes in the two variables. A notable example often used to illustrate this phenomenon is a positive relationship between the rate of drownings and a rise in ice cream sales. One could say that one causes the other, but actually, the outside factor causing both is the weather. Warmer weather in summer means that more people are likely to go swimming and more people are likely to buy ice creams.
- One variable causes the other: X causes Y or vice versa. A genuine causal relationship exists. The problem remains about which is the cause and which is the effect? It is possible that causation goes in the opposite direction, that Y causes X. The direction of time is one obvious factor to consider. There is a strong explanation as to why pulling the trigger causes a bullet being fired, for example: the trigger causes the hammer to strike the primer that ignites the gunpowder and the ensuing explosion propels the bullet.
Methods for investigating potential causal relationships
To establish a genuine causal relationship, several types of complementary evidence are required, ranging from simple to complex, common to rare. These range from:
- Basic correlational techniques that show how one factor (income for example) is related to other factors, without controlling for outside confounding factors; to
- Regression models that do control for these factors to eliminate alternative causal explanations and assess the relative strength of relationships; to
- Experimental studies, like randomised controlled experiments and natural experiments, that observe changes over time following interventions to give more confidence on the direction and strength of causality than the other methods.
In the social sciences, empirical tests of the causal mechanisms at the level of individual action are also necessary to understand the generative process linking the potential cause and effect. For example, lack of investment in children is commonly cited as a theory that may explain how poverty experienced in childhood causes poor outcomes later in life. To test this, a study would need to be designed to analyse how individuals make investment decisions with respect to their children’s development.
Correlational and regression techniques are the most common, while experimental studies are relatively scarce, particularly in New Zealand. We will primarily draw upon the first two types of evidence as they constitute the majority of the existing literature.
A worked example exploring correlation and regression
Different kinds of evidence also answer different questions, is early motherhood a direct cause of poverty? Researchers have attempted to untangle the causal relationship between early motherhood and poverty using data from the longitudinal birth cohort Christchurch Health and Development Study. Similar findings from regression models will be called upon in further sections, as will an analysis of how family changes trigger movements into and out of poverty, but for now this example is primarily here to help to illustrate what conclusions can be drawn from which data.
Association: Cause and effect?
- Association: Mothers who became parents before age twenty had significantly worse economic outcomes than women who weren’t mothers at the same age. These early mothers “worked fewer hours in paid employment; were more likely to be welfare dependent; had lower personal incomes; were more likely to experience economic hardship; were more likely to report that they did not have enough money for everyday needs; and were more likely to report that it was impossible for them to save.” These findings show there is a correlation between early motherhood and poverty.
But having a higher likelihood of being in poverty was not the only characteristic shared by those who had become mothers in their teenage years. When compared with women who were the same age but did not experience early motherhood, the authors find that:
early mothers were more likely to have been raised by a mother who was young, had no formal educational qualifications, and were a single parent, were more likely to be of Māori ethnicity, were more likely to have experienced sexual abuse before age 16, had lower IQ and academic achievement, had higher levels of childhood conduct problems and attention problems, were more likely to affiliate with deviant peers at age 14, and were more likely to have used alcohol by age 14.
These shared characteristics of family background and early life circumstances are covariates—other possible causal explanations for the relationship between early motherhood and poverty.
Sequence: Cause to effect?
- Sequence: To eliminate these alternative causal explanations, researchers use regression models with information from people’s lives over time to control for these potentially confounding factors: the “pre-pregnancy differences between early mothers and other women.”
There are two potential outcomes:
- If the relationship between early motherhood and economic outcomes disappears when these family background and early life circumstances are considered, then it’s likely that these factors are driving the relationship, rather than early motherhood itself—that is, that these family background and early life circumstances are likely causing both the early motherhood and the poverty. In this way, early motherhood is likely an indicator (or symptom/marker) of poverty rather than a likely cause of it. Indicators can potentially mask the true underlying causal factors.
- If the relationship between early motherhood and economic outcomes remains after these family background and early life circumstances are considered, it suggests that early motherhood does increase the risk of poor outcomes, regardless of the woman’s background and life circumstances before becoming a mother. We would be much more confident about the causal relationship between early motherhood and economic outcomes.
After controlling for potential confounds, the relationship “between early motherhood and all of the economic outcomes,” while diminished, was still significant. This suggests a potential causal relationship as many of the alternative explanations for the relationship were eliminated. Even after adjustment, early mothers were around “five times more likely to be welfare dependent at age 30,” around five times more likely to experience economic hardship, and earn $10,000 less per year than women who didn’t have children early in life. This “economic disadvantage … is still evident ten years after their entry into motherhood, suggesting long-term impacts rather than short-term setbacks.
Interventions: Effect from cause?
- Interventions: The statistical regression model used here is not an intervention-type experiment and cannot definitively establish causation. Instead, regressions are best understood as a tool to help rule out alternative explanations to provide additional confidence about potential causal relationships. We cannot, therefore, support a categorical claim that early motherhood causes poverty. A true experiment to draw more confident and robust causal conclusions would involve mothers being randomly assigned to have children then observing the effects. This has obvious ethical problems, and highlights the difficulty faced by social scientists in this area.
Explanations: How is effect caused?
- Explanations: The research used here has established an association but it cannot confirm the mechanism acting on this relationship. For this, a theory needs to be devised that links the potential cause and effect. In this case, the authors of the Christchurch study note “one possibility is that having a child before age 20 interferes with important life tasks that are being completed around this age, such as completing education or entering the job market. Failure to complete any of these tasks may limit an individual’s later opportunities for economic success.” This is, for now, a theory, and would also require additional empirical research that observed the level of individual/family action to support it.
It is important to use the right set of tools for the job. This section has shown how different research methods can uncover different aspects of reality so we can—to the extent that it is possible—assess the nature, strength and direction of causal relationships. The limitations of the available research methods have also been made apparent. Nevertheless, seeking to uncover potential causal relationships gives us “levers on reality, some basis for choosing how to act,” as political scientist W. Philips Shively writes, “coincidence without cause gives you no lever.” Making accurate distinctions between potential causes and mere indicators, for example, is crucial for targeting poverty-fighting policy well. We sometimes pull policy levers to no effect, and other times we simply pull the wrong levers entirely. A better understanding of the potential causal processes driving poverty will help guide better policy responses.
Our lives are constantly in flux. We experience predictable life-cycle changes like reaching adulthood, having children, and retiring one day, and, in an uncertain world, we also face unpredictable challenges like relationships falling apart or losing a job. Like ocean currents, broader economic and cultural forces also push and pull us about. It should therefore come as no surprise that our resources and needs fluctuate alongside these changes—predictable and unpredictable; favourable and unfavourable. An early illustration of income dynamics across the lifetime of an individual may be seen in Figure 3 below:
Individuals and families will often move up and down the income ladder, living on a relatively low income at one point, then moving to a higher income, and perhaps dipping back down. Needs change too as changes to family structure and size drive movements above and below the poverty line, as is mostly the case in Figure 3. Such movement is referred to as income mobility. The length of time a family spends on any one rung on the income ladder is referred to as persistence—the stay can be temporary, recurrent, or persistent.
Though it is individuals and families who experience income mobility, mobility is often thought about and discussed at an aggregate societal level. The images below in Figure 4 use data from a longitudinal survey called SoFIE to represent this movement across the entire income distribution over time in New Zealand between 2002 and 2009. These images can be difficult to interpret, but an analogy borrowed from Stephen Jenkins, LSE Professor of Economics and Social Policy, may help.
Imagine a multi-storey apartment building with ten floors, the poorest living in the basement (red), the richest in the penthouse (blue) and the rest dispersed in the middle floors in order of income. The left-hand side diagram shows this. While movement was tracked year-to-year, the right-hand side diagram shows movement among the levels after seven years. It answers the questions: “how much movement between floors is there…is there much turnover in the basement, and do basement dwellers ever reach the penthouse? Who moves the most and how far?”
Analysing the information illustrated in Figure 4 we see that after seven years, over a third of people are still living on the floor they started. Though there is a lot of movement, most movement is a short distance. From year to year, around 70 percent of New Zealanders stayed in the same level, or moved up or down one level. At the end of the period, 38 percent of New Zealanders on average (about 20 percent of families with children) were in the same level as when they started. The general trend was to move up or down a floor—using the stairs is more likely than needing to use the elevator. This holds across all income levels, for the rich and the poor alike, although the poor are relatively less mobile. These trends in mobility and persistence are broadly consistent across comparable OECD countries and reasonably independent of economic cycles.
Jenkins likens this dynamic to the elasticity of a rubber band:
Each person’s income fluctuates about a relatively fixed longer-term average—this value is a tether on the income scale to which people are attached by a rubber band. They may move away from the tether from one year to the next, but not too far because of the band holding them. And they tend to rebound back towards and around the tether over a period of several years. In the short-term some of the observed movement may simply be measurement error and, in the long term, the position of each person’s tether will move with secular [long-term] income growth or career developments. But, in addition, rubber bands will break if stretched too far by ‘shocks,’ leading to significant changes in relative income position.
Rather than a broader concern across the whole income distribution, in this paper we are concerned with the inhabitants of the bottom few floors of the income dynamics apartment building—the people who aren’t experiencing any lasting movement upward and are finding themselves tethered to those floors. We are especially concerned with the inhabitants whose stay in the basement persists for many years.
While we often have a pre-determined idea of what “the poor” are like, they are not a fixed group. A family struggling with low income that shows up in the headline statistics one year is not necessarily in the same financial position the next year. The evidence shows that people’s incomes—all across the financial spectrum—shift significantly over time. The term poverty can be misleading, as “many more people are touched by poverty over time than are poor in any given year;” more than we might assume.
Consider an illustration: a hospital ward with ten beds. Imagine this ward is the bottom floor of the building in the above example. If we were to walk around the ward and ask how long patients had been there for, and the majority responded that they’d been there for over a year, we could reasonably assume that the “average patient” was in the ward for over a year. The bed in the corner with someone staying for just a few days would seem like a minority. But if we repeated our walk around the ward every week for a year, we’d meet a new person each week in the bed in the corner. If, for example, nine patients out of ten were there for the whole year with a new person each week in the tenth bed, there would be fifty-two short-term stays and nine long-term. The minority at a point-in-time become the majority over a period of time.
The evidence shows that poverty works in a similar way. While a snapshot of data at one point in time will show what proportion of people are in poverty, it says nothing about how long they’ve been there. Poverty rates do not distinguish between those passing through, those dipping in and out, and those languishing there for many years. To account for these different types of poverty, longitudinal research shows that poverty can be classified into three categories: transient, recurrent, or persistent.
Over time, most people who fall into poverty are there for a short spell—these are the transient poor. At any point in time, around a third of New Zealand households with low income are there temporarily. This means that more people are affected by poverty over a period of years than are poor for a single year.
In New Zealand, around a quarter of New Zealanders have a low income for any particular year, but over the course of seven years around half of the population— more than double those experiencing poverty at any one time—experienced poverty for at least one year. This makes sense considering times when people are studying, between jobs, or recovering from being sick or having a child.
Most people who escape poverty do not fall back into poverty for a long time, but a “non-negligible fraction” do—these are the recurrent poor. The data behind Jenkins’ rubber band illustration explains this to some extent. Families just below the poverty line, for example, may rise up for a time but on average they will tend to “rebound” back towards were they started. This results in frequent transitions across the poverty “line.” Given median income measures, it is also possible that these transitions are due to changes in the median income like those experienced during a recession.
While, across society, most of those who started out on lower incomes will end up better off in real terms over their lifetimes, sadly, there are many who remain mired in poverty for longer spells—often referred to as persistent poverty.
According to SoFIE data, 16 percent of New Zealanders experienced persistent poverty, defined as having a low income for at least five of the seven years surveyed. An alternative way of measuring persistent poverty that averages incomes over a number of years (to allow for savings/debt cycles) finds that around 70 percent of those experiencing point-in-time poverty have or will be there persistently; 80 percent of children and Māori experiencing point-in-time poverty have or will be there persistently. Using this measure, around 11 percent of all New Zealanders, 18 percent of children and 19 percent of Māori experienced poverty persistently.
While persistent poverty affects a relatively small share of the population, the effects on people’s lives are significant. The longer a family is in poverty, the more likely it is that they will experience greater levels of hardship and more severe poverty. Those “that had low income for seven years on average were more than three times more likely to report being in hardship than those people who had low income in one year.”
Poverty now also drives poverty in the future. In what has been called the “state dependence” effect, being poor one year raises the chances of being poor the next, even when other factors have been controlled for using a multivariate model. New Zealand data suggests that a family with low income one year is 65 percent likely to remain on low income the following year. International evidence suggests that families are four times more likely to enter poverty if they have been in poverty in the past. Not only are families “who have experienced poverty in the past … more risk of entering poverty than those who have not been in poverty,” international evidence suggests that the longer they stay there the less likely it is that they’ll escape. This is why poverty is often described as a “trap.”
Cumulative impact of poverty
The “cumulative impact” of persistent poverty can scar deeply, with effects shown even into the next generation, particularly when a family experiences persistent poverty when their children are young. Children in persistently poor families suffer more and are more likely to be poor themselves in the future.
It is these long-term consequences of persistent poverty that lead us to concentrate the remainder of this paper on this group. In turning to this group now, we are interested in both current poverty—that is, what keeps poor families poor within lifetimes—and future poverty—that is, what makes poor children grow up to be poor adults across generations.
This is an extract from Kieran’s research series “The Heart of Poverty | Uncovering Pathways into and out of Disadvantage in New Zealand” Discussion Paper. (Released 2016)
 Susan E. Mayer, The influence of parental income on children’s outcomes, (Ministry of Social Development, 2002), 6.
 Jonathan Boston and Simon Chapple, Child Poverty in New Zealand (2014), 48, 262-263. See also Expert Advisory Group on Solutions to Child Poverty (EAG), Working Paper no.2: Life course Effects on Childhood Poverty (Office of the Children’s Commisioner, 2012), 2 for another comprehensive list of studies.
 For a discussion on what statistics can say about causation, see Paul Holland (1986) “Statistics and Causal Inference,” Journal of the American Statistical Association, 81, no. 396, 945-60.
 Shively, The craft of political research, 77-78.
 Ariel Kalil, “Family resilience and good child outcomes: A review of literature” (MSD, 2003): 38.
 Shively, The craft of political research, 78.
 John Goldthorpe “Causation, statistics, and sociology.” European sociological review 17, no. 1 (2001): 1-20. We will look at these theories below. For an excellent example of research that tests a causal theory at the individual level (in this case the hypothesis that income inequality causes poor outcomes through “status anxiety”), see Richard Layte & Chistopher T. Whelan, “Who feels inferior? A test of the status anxiety hypothesis of social inequalities in health.” European Sociological Review 30, no. 4 (2014): 525-535.
 Randomised controlled Experiments are inherently difficult to implement in New Zealand due to relatively small geographical size and population.
 For an overview of recent experimental evidence, see Boston and Chapple, Child Poverty in New Zealand, 51-54.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes: Findings from a 30-Year Longitudinal Study,” Journal of Research on Adolescence 25, no. 1 (2015): 163-172.
 There was insufficient data to track the impact of early fatherhood.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 170.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 168 (P values ommitted from citation). A wide range of additional covariate factors were added to the model but were found to be insignificant, suggesting that the family background and early life circumstances were relatively comprehensive. Different cut-off points (ages 19, 22 and 25) were also tested but didn’t alter the results significantly.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 163.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 171.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 169-170.
 Gibb et al., “Early Motherhood and Long-Term Economic Outcomes,” 171.
 Kalil, “Family resilience and good child outcomes: A review of literature,” 38.
 Kristin Moore et al., “Age at first child- birth and later poverty,” Journal of Research on Adolescence, 3, 393–422 cited in Gibb et al (2014) “Early Motherhood and Long-Term Economic Outcomes,” 171.
 Shively, The craft of political research, 76
 The notion of equivalised income is important to grasp. When comparing families of different sizes and compositions the income needs to be adjusted because larger families have greater needs than smaller families (and also also greater economies of scale). If incomes were not equivalised, relative comparisons would not be meaningful. See Perry, Household incomes in New Zealand, 9-11 for more on equivalisation.
 For more information on SoFIE, see Perry, Household incomes in New Zealand, 168.
Stephen P. Jenkins, Changing fortunes: income mobility and poverty dynamics in Britain (OUP, 2011).
 Treasury believes that this mobility seems unlikely to be related to “retirements or entry from education as the pattern was essentially the same for the 25 to 55 agegroup,” a population that avoids those life changes. New Zealand Treasury, A descriptive analysis of income and deprivation in New Zealand, 8.
 Jenkins, Changing fortunes.
 The income measure used here is “gross equivalized household income before tax but after benefits have been paid (including tax credits).” New Zealand Treasury, A descriptive analysis of income and deprivation in New Zealand, 17
 It cannot be assumed that they did not move during that period, however. Kristie Carter, Penny Mok and Trinh Le, Income Mobility in New Zealand: A Descriptive Analysis: New Zealand Treasury Working Paper 14/15 (2014), 15-16. For comparable Australian figures, see Roger Wilkins ed., Families, Incomes and Jobs, Volume 9: A Statistical Report on Waves 1 to 11 of the HILDA Survey (University of Melbourne, 2011).
 Longer studies of this nature (such as the 16 year UK study) have generally found that mobility is reasonably constant through the economic cycle. See New Zealand Treasury, A descriptive analysis of income and deprivation in New Zealand, 17. For a comparison with Australia and the United Kingdom, see New Zealand Treasury, A descriptive analysis of income and deprivation in New Zealand, 21. Where differences in persistence arise across countries, research suggests that demographic factors play a role—having more families with children, for example—but “differences in the poverty-generating process” are behind most of the variation. Jenkins, Changing fortunes. 329. Countries with higher poverty rates using static measures also tend to have higher rates of persistent and recurrent poverty using dynamic measures.” See also OECD, Doing Better for Children, 152-158; Ballantyne et al, Movements Into and Out of Child Poverty in New Zealand 23-28.
 Stephen P. Jenkins, Richard Berthoud and Jonathan Burton, Income and poverty: the rubber band theory (Institute for Social and Economic Research, 2008): 2 cited in Jenkins, Changing Fortunes, 361.
 For a philosophical argument for focussing our collective efforts on reducing poverty rather than broader inequality, see Harry Frankfurt, On Inequality (Princeton University Press, 2015).
 John Hills, Good times, bad times: The welfare myth of them and us (Policy Press, 2014). See also Jenkins, Changing fortunes, 363, Noel Smith, and Sue Middleton, A review of poverty dynamics research in the UK (Joseph Rowntree Foundation, 2007), 1.
 This illustration is taken from pioneering work on poverty dynamics by Mary Jo Bane and David T. Ellwood, Slipping into and out of poverty: The dynamics of spells (1983), 11-12.
 Bruce Bradbury, Stephen P. Jenkins and John Micklewright, eds. The dynamics of child poverty in industrialised countries (Cambridge University Press, 2001), 1.
 Benefit receipt operates in a similar manner. See Centre for Social Research and Evaluation, Who uses the benefit system and for how long? (MSD, 2010), SusanMorton et al., Growing Up in New Zealand, Now We Are Two: Describing our first 1000 days (2014), 43.
 OECD, Growing Unequal (2008) 157. See also Ruud Muffels, Didier Fouarge and Ronald Dekker, Longitudinal Poverty and Income Inequality: A Comparative Panel Study for the Netherlands, Germany and the UK, OSA Working Paper WP2000-6 (2000)
 New Zealand Treasury, Improving outcomes for children – Initial Views on Medium-term Policy Directions, Report to the Ministerial Committee on Poverty, (2013), 9
 Jenkins, Changing fortunes, 363.
 New Zealand Treasury, A descriptive analysis of income and deprivation in New Zealand, 10.
 OECD (2008), Growing Unequal, 156.
 Jenkins, Changing fortunes, 326.
 Carter, Mok and Le, Income Mobility in New Zealand: A Descriptive Analysis, 18. See also Perry, Household incomes in New Zealand, 167-180.
 This is referred to as “chronic poverty” in the literature. Perry, Household incomes in New Zealand, 178.
 Using 60% of the median, after housing costs (AHC). OECD, Economic Surveys: New Zealand 2015 (2015), 119. Boston and Chapple note that these figures mayunderestimate the persistence of poverty, as SoFIE uses net rather than gross income data, which are “likely to be less volatile.” Child Poverty in New Zealand, 45.
 New Zealand Treasury, Improving Outcomes for Children, 9.
 Jenkins, Changing fortunes, 355. To clarify, the “state” here refers to away of being/status, rather than the government.
 Kristie Carter and Fiona Imlach Gunasekara, Dynamics of Income and Deprivation in New Zealand, 2002-2009: A Descriptive analysis of the Survey of Family, Income and Employment (SoFIE) (University of Otago, 2012), 9.
 Department of Work and Pensions, An evidence review of the drivers of child poverty for families in poverty now and for poor children growing up to be poor adults (2014), 98.
 Smith and Middleton, A review of poverty dynamics research in the UK, 44. See also Bane and Ellwood, Slipping into and out of poverty: The dynamics of spells,OECD, Growing Unequal, 165.
 For more on the cumulative impact of low income on deprivation, see Perry, The material wellbeing of New Zealand households, 56. For the intergenerational impact, see Greg J. Duncan and Katherine Magnuson, “The importance of poverty early in childhood.” Policy Quarterly 9, no. 2 (2013): 12-17.
 UK Department of Work and Pensions, An evidence review of the drivers of child poverty for families in poverty now and for poor children growing up to be poor adults, 9.
 This twin focus on poverty persisting in families now and in the future is based on and draws heavily work by the UK Department of Work and Pensions: An evidence review of the drivers of child poverty for families in poverty now and for poor children growing up to be poor adults. Comparable New Zealand data is limited, however, the levels and trends of income persistence and mobility are broadly comparable to other international studies. Britain’s child poverty dynamics, at least within lifetimes, were also found to have the closest resemblance to New Zealand’s when compared to a host of European countries. See: Carter, Mok and Le, Income Mobility in New Zealand: A Descriptive Analysis, 1; and Suzie Ballantyne et al., Movements Into and Out of Child Poverty in New Zealand: Results from the Linked Income Supplement (Motu, 2003), 28. The authors cautions that this likeness could be partly attributed to similarities in weekly survey methodology shared between New Zealand and Britain. Longitudinal data will be used where possible (sometimes only point-in-time data is available).