Measuring Childhood Exposure to Neighbourhood Deprivation at the Macro- and Micro-level in Aotearoa New Zealand

Neighbourhood effects research has benefited from the application of sequence analysis which, together with cluster analysis, identifies the main temporal patterns of exposure to residential contexts experienced by different groups of people, such as children. However, given that this is a relatively new approach to measuring exposure to neighbourhood deprivation, studies that have utilised sequence analysis to model residential trajectories and test for neighbourhood effects do not contextualise these population-level findings at the individual-level. The current study sought to investigate the patterns of exposure to neighbourhood deprivation experienced by children in Aotearoa New Zealand over the first eight years of life by utilising two different methodological approaches: at the macro-level, the results of the sequence and cluster analysis suggest that in general, children experienced little neighbourhood mobility; at the micro-level, children experienced greater levels of movement between different levels of neighbourhood deprivation in middle childhood, compared to early childhood, while children in the least and most deprived neighbourhoods experienced less mobility than their peers. Together, these findings provide a comprehensive description of the ways in which children are exposed to different residential contexts over time and advance our understandings of how to document these experiences effectively within quantitative research.


Introduction
Neighbourhood effects research has found that there is an association between the neighbourhood in which a child grows up in and their later health and developmental outcomes (DeLuca & Jang-Trettien, 2020;van Ham & Manley, 2010). A neighbourhood can be therapeutic for its residents and promote wellbeing, or pathogenic and compromise the health of the community (Pearce, 2018), depending on the different social, environmental, geographical, and institutional mechanisms at play (Galster, 2012). Neighbourhood effects research usually focuses on the impacts of area-level disadvantage, under the premise that current and cumulative exposure to high levels of neighbourhood deprivation, and the physical, cultural, and social environments that this deprivation embodies, is detrimental for individuals, particularly when this exposure is during childhood. Sharkey and Faber (2014) noted that this line of research can ultimately be reduced to a single question: 'do neighbourhoods matter?'. There is concern that the framing of this question perpetuates a dichotomous understanding of neighbourhood effects and trivialises the research field (Sharkey & Faber, 2014). Indeed, neighbourhood effects are complex -a consequence of the dynamic interaction between neighbourhood characteristics, the timing, duration, and sequence of exposure to residential contexts, the family, and the individual themselves (Harding et al., 2010). These components must be understood both individually and as a collective to effectively account for the heterogeneity in neighbourhood effects experienced by different groups of people. Thus, the exploration of the temporal elements of exposure to neighbourhood deprivation may benefit interventions aimed at reducing negative neighbourhood effects across the lifespan.
Studies of neighbourhood effects typically utilise a single point-in-time measure of residential context in their analyses (e.g., Duncan et al., 1999;Sykes & Kuyper, 2009). Using cross-sectional measures of neighbourhood deprivation, however, assumes that one's local environment remains consistent over time (Kleinepier & van Ham, 2017), despite potential changes in exposure caused by residential mobility or change in neighbourhood context or composition. Rates of residential mobility vary by country, with Australia and New Zealand having some of the highest levels of mobility among developed countries (United Nations, 2013). Furthermore, some of the highest rates of residential mobility are experienced by the youngest sector of society (0-4 years; Jian & Dasgupta, 2017; Nathan et al., 2019). Van Ham and Manley (2012) identified the consideration of people's residential histories as one of ten methodological challenges and recommendations for future neighbourhood effects research, while Pearce (2018) noted that inaccurate measures of people's environmental exposures were problematic for the quantitative study of neighbourhood effects. The recent application of life course theory has provided a longitudinal approach through which to capture people's residential trajectories more accurately. Life course theory proposes that human development is a lifelong process and that earlier experiences have lasting impacts on later life chances (Elder et al., 2003;Hagedoorn & Helbich, 2021;Mayer, 2009). Studies of human development must consider people's lives holistically and acknowledge the social, historical, and cultural dynamics shaping experiences, processes, and outcomes. Applied to neighbourhood effects research, life course theory promotes the analysis of longitudinal data to investigate how different groups of people are exposed to neighbourhood deprivation in childhood and how this results in the heterogeneity of neighbourhood effects.
Drawing on life course theory, sequence analysis can be used to model the timing, duration, and sequence of exposure to residential contexts across the lifespan by combining multiple timepoints of data into a single sequence (Abbott, 1990). Cluster analysis is then used to identify the main temporal patterns of exposure to neighbourhood deprivation contained in the sequences, thus providing macro-level findings about people's residential experiences. For example, Kleinepier and colleagues (2018) applied sequence analysis to capture children's exposure to neighbourhood deprivation over the first 15 years of life. Six clusters were identified, including three characterised by stability (consistent deprivation, consistent middle-income, and consistent affluence) and three characterised by change (early deprivation, adolescent deprivation, and early affluence).  utilised a similar dataset and identified seven clusters. It should be noted that these clusters are somewhat determined by the unit of measure used to capture neighbourhood deprivation: in both cases neighbourhoods were firstly sorted into quintiles based on neighbourhood income and then collapsed into tertiles (termed 'deprived', 'middle-income', and 'affluent' neighbourhoods). Where neighbourhood deprivation is captured in smaller measurement units there is a correspondingly greater number of possible clusters that can be characterised by either stability or change. Effective labelling and descriptions of the identified clusters can assist in the interpretation and comparisons of findings across studies.
Sequence analysis has been used to model residential trajectories in various countries, such as Canada (Letarte et al., 2021) and the Netherlands , and for different groups of people, such as migrants (Vogiazides & Chihaya, 2020) and children . However, sequence analysis has not yet been conducted to identify the main temporal patterns of exposure to neighbourhood deprivation commonly experienced in Aotearoa 1 New Zealand (NZ). In their systematic review of neighbourhood effects on early childhood development Minh et al. (2017) noted that findings were not generalisable across places. Indeed, against expectations and existing international research (e.g., Guagliardo et al., 2004), Pearce and colleagues (2007) found that access to healthrelated community resources, such as the availability of a doctor, were greater in more deprived neighbourhoods compared to less deprived neighbourhoods, within the Aotearoa NZ context. Thus, it is important to model exposure to neighbourhood deprivation in culturally varied contexts to provide international points of reference for studies investigating the residential trajectories of different groups of people, and to identify both universal and context-specific causal mechanisms that underpin neighbourhood effects.
The application of sequence and cluster analysis for the study of neighbourhood effects is typically data-driven in its identification of clusters (see as an exception Letarte et al., 2021). These clusters are interpreted and labelled by the researcher to provide a description of the macro-level trends in residential trajectories. However, it is also important to contextualise these patterns at a more granular (i.e., micro-) level. Even within a cluster, for example, it is unlikely that all people will experience the exact same sequence of exposure to neighbourhood deprivation. Analysing data at the individual-level, such as investigating how much movement between levels of neighbourhood deprivation is experienced, can help to shed light on whether there are micro-level trends that are representative of people's timepoint-to-timepoint residential experiences. Furthermore, given that results of sequence analysis become either the independent (e.g.,  or dependent (e.g.,  variable of interest in studies of neighbourhood effects, there is little exploration about how macro-and micro-level analyses of residential trajectories may provide complementary or diverging insights about the temporal patterns of exposure to various residential contexts experienced by different groups of people. Neighbourhood deprivation data is often included as a variable in studies of childhood health and wellbeing, either as the primary variable of interest (e.g., studies of neighbourhood effects;  or as a covariate (e.g., Russell et al., 2022). Thus, a consideration of different methodological approaches to the measurement and analysis of such data is important because these researcher decisions determine the samples used for analysis, such as the identification of 'high deprivation' samples for studies of resilience (e.g., Ball et al., 2013). Accordingly, the current study will explore exposure to neighbourhood deprivation at both the macroand micro-level to gain a comprehensive understanding of how children in Aotearoa NZ are exposed to neighbourhood deprivation over time, as well as to identify the different contributions to knowledge these two levels of analysis can make to this research field.
Lastly, childhood is a sensitive period of development in which the social, economic, cultural, and built environment lay the foundations for future health and wellbeing outcomes (Minh et al., 2017). Moreover, children often spend more time within their local environment than adults who may commute outside of their residential context for work. This likely strengthens neighbourhood effects by increasing the amount of exposure to the local environment at a particularly sensitive period of development and therefore shaping children's developmental trajectories. For example, Ivory and colleagues (2015) found that the association between the built environment and physical activity was stronger for people who were more reliant on their residential neighbourhood, such as those with restricted car access. Children similarly rely on adults to access locations outside of their local environment. In Aotearoa NZ, 95% of children in elementary school attend a state institution (Education Counts, n.d.), which is usually located within the residential neighbourhood. This indicates that during the working week, most children are interacting with people and places from within their local environment. Differences in residential experiences between sociodemographic groups, including age groups, are reflected in the heterogeneity of neighbourhood effects and require further research to identify the causal mechanisms of such effects (Sharkey & Faber, 2014). Given the assertion that exposure to high levels of neighbourhood deprivation in childhood can be detrimental for individuals, documenting children's residential experiences provides important information about the contexts in which children are developing within and how they impact on wellbeing across the lifespan.
The investigation of neighbourhood effects using a life course approach is in its infancy (Jivraj et al., 2019). Accurately accounting for people's residential trajectories during childhood is an important component of this line of research, with sequence analysis providing one approach to identifying the timing, duration, and sequence of exposure to neighbourhood deprivation experienced at the population-level. Furthermore, the concurrent exploration of exposure to neighbourhood deprivation at the micro-level will provide a detailed description of the residential experiences of children growing up in Aotearoa NZ. Concurrent analysis of residential trajectories at the macro-and micro-level also offers a unique opportunity to compare and contrast the conclusions drawn about the nature of children's exposure to neighbourhood deprivation derived from two different methodological approaches: the populationlevel sequence and cluster analysis, and individual-level analyses about the quantity and direction of movement between different residential contexts over time. Given the outlined research gaps and the opportunities identified, the research questions are: 1. What are the main patterns of exposure to neighbourhood deprivation experienced by children in Aotearoa NZ? 2. How do children move between different levels of neighbourhood deprivation? 3. What are the similarities and differences in the findings produced by analysing exposure to neighbourhood deprivation at the macro-and micro-level?

Data
The Growing Up in New Zealand (GUiNZ) study is a longitudinal child cohort study designed to produce policy-relevant research for multiple areas of child development in Aotearoa NZ (Morton et al., 2014). The study recruited 6,846 children via their mothers who were due to give birth between April 2009 and March 2010. The cohort was recruited to be representative of all births during this period (Morton et al., 2013). Ethical approval for the GUiNZ study was obtained from the Ministry of Health "Northern Y" Regional Ethics Committee. The GUiNZ study has a high retention rate of child participants over the first 8 years of life, with 76% of the baseline cohort captured at every major data collection wave (DCW): antenatal, 9-months, 2-years, and 54-months (Morton et al., 2020). Families of the Growing Up in New Zealand children are supported to participate in all DCWs but are permitted to 'skip' DCWs where participation is not possible. The current study used mother report measures collected during the antenatal DCW, as well as five other timepoints over the first 8 years of their child(ren)'s life: 9-months, 2-years, 54-months, 72-months, and 8-years. Neighbourhood deprivation was collected at each of these DCWs, while mother sociodemographic characteristics were collected antenatally. Although all measures used in the current study were reported by the mother and were reflective of the mothers' circumstances, the analyses were designed to be child-centred, meaning that exposure to neighbourhood deprivation was interpreted in terms of the child experience and makes the child the individual unit of analysis. However, mother sociodemographic characteristics were used to contextualise children's experiences of neighbourhood deprivation.

Neighbourhood Deprivation
NZDep2006 and NZDep2013 are small-area indices of relative socioeconomic deprivation derived using census data (Salmond & Crampton, 2012). These small areas have a median of 90 residents. The levels of deprivation are calculated as deciles, where a value of 1 means that the relevant area is in the least deprived 10% of all small areas in Aotearoa NZ (Salmond et al., 2007). This score of socioeconomic deprivation is the first principal component derived from the principal component analysis of nine variables that capture eight dimensions of deprivation, including income, living space, and communication, collated at the area-level. These scores explained 55% (NZDep2006) and 62% (NZDep2013) of the overall variance (Atkinson et al., 2014). Given that all nine variables reflect a lack of something, the distribution of NZDep scores reflects experiences ranging from the lowest levels of deprivation (Decile 1) to the highest levels of deprivation (Decile 10), rather than from affluence to deprivation. GUiNZ collected NZDep2006 data up until the 54-month DCW when the updated NZDep2013 measure was implemented. The six NZDep measures analysed in the current study were highly correlated over time (see Table 1). Across the six timepoints, 32% of cases had at least one missing NZDep value. To ensure that the sample was as representative as possible, cases with fewer than three missing values had their NZDep data multiply imputed through prediction based on a multinomial regression model using the surrounding NZDep values. Five possible values for the missing data were calculated, resulting in five unique datasets. Sequence and cluster analysis were conducted on each dataset separately and the results compared for robustness (see Appendix A). This process increased the sample size by 26% to 5,853 (hereafter referred to as the 'sample for analysis') and was conducted using the R package (R Core Team, 2019) seqimpute (Berchtold, 2021). All results presented in this study have been calculated using the sample for analysis.

Mother Sociodemographic Characteristics
Six sociodemographic characteristics of mothers collected at the antenatal DCW were used to describe the make-up of each cluster and quintile at each timepoint: ethnicity, education, employment status, rurality, residential mobility, and neighbourhood deprivation. The mother participants were asked to identify all ethnic groups that they belonged to. For those who nominated more than one ethnic group, administrative-prioritisation was used to identify a single ethnic identity. Administrativeprioritisation is implemented by the researcher, based on a predetermined hierarchy (Yao et al., 2022). In Aotearoa NZ this hierarchy was determined by the former Department of Statistics (1993). Its prioritisation of the Māori ethnic identity recognises the country's commitment to upholding the principles of Te Tiriti o Waitangi (the Treaty of Waitangi), while the subsequent prioritisation of the Pacific and Asian ethnic identities, followed by other ethnic groups (excluding NZ European), reflects the relative size of these ethnic groups, and thus acknowledges them as minority groups (Yao et al., 2022). As the dominant ethnic group, the NZ European identity was the lowest prioritised group. Mothers were also asked to report their highest educational qualification, employment status, and how many times they had moved to a new house over the previous five years, while rurality (i.e., whether they lived in a rural or urban area) was derived from supplied residential data. Lastly, the average antenatal NZDep quintile was calculated for each cluster and quintile.

Data Analysis
The current study sought to identify the main patterns of exposure to neighbourhood deprivation during childhood and contextualise these findings by exploring the nature of movement between residential contexts over time. This section outlines the method of data analysis used to combine the cross-sectional measures of neighbourhood deprivation into sequences and cluster these sequences into meaningful groups to produce the macro-level findings. The section then outlines how the quantity and direction of transitions between different levels of neighbourhood deprivation of individuals were calculated to produce the micro-level findings.

Sequence and Cluster Analysis
Sequence analysis was conducted to capture the timing, duration, and sequence of exposure to neighbourhood deprivation during childhood. Sequence analysis converted the NZDep quintiles (i.e., states) into characters which were used to create sequences that described a child's residential trajectory. Cluster analysis was then performed to develop a typology of children's neighbourhood deprivation trajectories. Firstly, optimal matching distances between trajectories were calculated by using an insertion/deletion cost of 1 and a substitution cost of the inverse of transition frequencies, aligning with similar studies of neighbourhood deprivation (e.g., Letarte et al., 2021). These values determine how much it 'costs' to transform from one trajectory to another. Secondly, partitioning around medoids was used to group trajectories together based on their optimal matching distances. This process identifies medoids, objects that are representative of the structure of the data, and assigns each sequence to the nearest medoid to produce clusters that maximise the between-cluster variability and minimise the within-cluster variability (Kaufman & Rousseeuw, 1990). A range of cluster solutions (i.e., number of clusters) was tested (2:30 cuts) and assessed using the average silhouette width criterion, which measures how well the sequences align with the nominated medoids (Kaufman & Rousseeuw, 1990). This process was conducted on each of the five imputed datasets. Following robustness tests, findings related to the sequence and cluster aollected at the antenatal Dnalysis were reported for the first dataset only. All analyses related to the sequence and cluster analyses were performed using the R package (R Core Team, 2019) TraMineR (Gabadinho et al., 2009).
Mother sociodemographic characteristics were used to describe cluster membership. Chi-square tests were used to test whether each of the sociodemographic characteristics (e.g., education) were associated with cluster membership. The same characteristics were also used to describe quintile membership at each of the six timepoints. The individual measures of neighbourhood deprivation were then used to contextualise the clusters produced by the sequence analysis by calculating how similar or different children's cross-sectional experiences of neighbourhood deprivation were compared to the cluster that they were assigned to (i.e., quintile-cluster discontinuity). At each timepoint, the children whose quintile did not align with their assigned cluster were identified and calculated as a proportion of the sample for analysis.

Analysing the Nature of Movement Between Levels of Neighbourhood Deprivation
To determine the quantity of movement between levels of neighbourhood deprivation experienced by children and the direction of this movement, the proportion of children moving to a different quintile between the six sequential timepoints (i.e., five transition points) was calculated (i.e., quintile-quintile discontinuity). This movement was further broken down by the quintile that children were moving from (i.e., base quintile). Chi-square tests were used to test whether there was an association between base quintile and likelihood of children moving out of that quintile.
Finally, to establish the direction of children's movement between levels of neighbourhood deprivation, each movement was coded as either + 1 (i.e., moving into less deprived areas) or -1 (i.e., moving into more deprived areas). At each transition point, the proportion of upward and downward mobility was averaged to determine whether there was an overall trend in direction. This was graphed to illustrate the net direction of movement across the five transition points.

Results
The current study sought to identify the main patterns of neighbourhood deprivation that children are exposed to in Aotearoa NZ, as well as to investigate the quantity and nature of movement between quintiles experienced at the individual-level. Firstly, this section presents the five clusters identified by the sequence and cluster analysis, including a sociodemographic description of these clusters and their alignment with cross-sectional measures of neighbourhood deprivation. Secondly, the micro-level analyses, focussing on the nature of the movement between quintiles at each of the five transition points, are presented to contextualise the macro-level findings.

Neighbourhood Deprivation Clusters
Sequence analysis was conducted to capture the timing, duration, and sequence of children's exposure to neighbourhood deprivation over the first 8 years of life. Cluster analysis was then used to identify subtypes of trajectories which reflected similar patterns of exposure to neighbourhood deprivation. Several cluster solutions were tested (2:30 cuts), of which the 5-cluster solution was determined to be optimal for each of the five imputed datasets (average silhouette widths ranged from 0.419 to 0.422 across the five imputed datasets). The five clusters generally aligned with the five quintiles of the NZDep2006 and NZDep2013 measures utilised in the current study (see Fig. 1). That is, there was little movement between levels of neighbourhood deprivation, suggesting that most children were consistently exposed to the same or very similar levels of neighbourhood deprivation over time. The five clusters were, therefore, labelled according to the quintile that they most represented: Cluster 1 (Quintile 1), Cluster 2 (Quintile 2), Cluster 3 (Quintile 3), Cluster 4 (Quintile 4), and Cluster 5 (Quintile 5). Mother sociodemographic characteristics were used to describe the composition of each cluster (see Table 2). Regarding ethnicity, children with NZ European mothers were less frequently living in more deprived clusters, whereas the opposite was true for the Māori and Pacific ethnic identities. Children with Asian mothers were most represented in Cluster 4, while the representation of the Other ethnic identity remained fairly consistent across clusters, although this group represented less than 5% of the sample for analysis. Of particular note, while less than 2% of children in Cluster 1 had mothers who identified as Pacific, nearly 35% of children in Cluster 5 did so, suggesting that there can be large inequities in the residential contexts that children are growing up in, in relation to their mothers' ethnic identities. The significant chi-square test result supports this interpretation (X 2 (16, 5846) = 1630.40, p < .001).
Related to education and employment status, children whose mothers had higher levels of education or who were employed during the antenatal DCW had higher levels of representation among the less deprived clusters. For example, 30.4% of children in Cluster 1 had mothers who had a Bachelor's degree, whereas just 10% of children in Cluster 5 had mothers who did so. Furthermore, children whose mothers lived in rural areas were more frequently living in less deprived clusters, but only 7.5% of the sample for analysis were living in rural areas. For all three of these sociodemographic characteristics, the chi-square test result indicated that there was an association between the characteristic and cluster membership, whereas residential mobility was not (X 2 (20, 5834) = 29.75, p = .07). Lastly, the average NZDep quintile value (measured at the antenatal DCW) increased as the clusters represented greater levels of neighbourhood deprivation. This suggests that labelling the clusters in alignment with the quintile that they most represented was an accurate description of the residential contexts that children in each of the clusters were experiencing.

Alignment Between Cluster Membership and cross-sectional Measures of Neighbourhood Deprivation
Given that the clusters were characterised by stability rather than change, it was possible to compare cluster membership, which spanned the entire childhood period, with the quintiles of neighbourhood deprivation experienced at each DCW. Levels of discontinuity were calculated to identify the timepoints where there were greater levels of misalignment between the cluster that children were assigned to and their actual point-in-time exposure to neighbourhood deprivation. Figure 2 illustrates that there were higher levels of quintile-cluster discontinuity in the middle childhood years, compared to the early childhood years. In particular, the highest levels of discontinuity were experienced at the 54-month and 8-year DCWs. The actual percentage of misalignment differed by cluster, with children in the most extreme clusters (Cluster 1 and Cluster 5) experiencing lower levels of discontinuity over time.

Neighbourhood Deprivation Quintiles
At each of the six timepoints children could be exposed to one of five levels of neighbourhood deprivation. Mother sociodemographic characteristics were used to describe the composition of the quintiles at each DCW. Table 3 presents the findings for the 8-year DCW and show that children were exposed to different levels of neighbourhood deprivation at different frequencies depending on their mothers' ethnicity, educational history, employment status, rurality, and level of residential mobility. While children could be in different quintiles across the six timepoints, the mother sociodemographic characteristics of the quintiles remained consistent over time, with only residential mobility returning variable chi-square test results (see Appendix B).

The Nature of Movement Between the cross-sectional Measures of Neighbourhood Deprivation
In addition to identifying the main patterns of exposure to neighbourhood deprivation experienced at the population-level, the current study sought to explore the amount of movement in and out of different levels of neighbourhood deprivation across the five transition points and whether this movement was characterised by a directional trend. That is, whether children were more likely to experience upward (i.e., moving into less deprived areas) or downward mobility (i.e., moving into more deprived areas). Indeed, while Fig. 3 shows that large portions of the sample remained in the same quintile over time, 77% of children moved to a different quintile at least once over the six DCWs, seemingly with increasing frequency after the 2-year DCW.

Quintile-quintile Discontinuity
To investigate how much movement children were experiencing, and between what timepoints this movement was being experienced, the level of quintile-quintile discontinuity was calculated by determining whether a child stayed in the same quintile, or not, between one DCW and the next. For each of the five transition points, the number of children who moved between different quintiles was calculated as a proportion of the sample for analysis (see Fig. 4). This set of analyses was conducted irrespective of cluster membership. The lowest levels of movement were  found between the antenatal and 9-month DCWs (16.9%). This percentage increased to over 50% between 2-years and 54-months. While the misalignment between the 2-year and 54-month DCWs was to be expected (52.5% discontinuity) due to the introduction of the NZDep2013 measure, the sustained high level of discontinuity at the transition to 72-months (47.2%) and 8-years (49.0%) was not. When this movement was analysed by base quintile (i.e., where children were moving from), floor and ceiling effects caused by being in the most and least deprived quintiles were observed (see Table 4). That is, there were greater levels of movement experienced by children who moved from the middle three quintiles in which children were able to move either into a more deprived or less deprived quintile. The

Fig. 3 Sankey plot of NZDep quintile transitions between data collection waves
significant chi-square test results at each transition point indicate that the likelihood of moving between different quintiles was associated with the quintile that children were moving from. In alignment with the results presented in Fig. 4, the proportion of quintile-quintile discontinuity increased for all base quintiles over time. For example, for children moving from Quintile 4, the percentage of those moving to another quintile more than tripled from 19.8% at the first transition point (antenatal DCW to 9-month DCW) to 62.2% at fifth and final transition point (72-month DCW to 8-year DCW).

Direction of Movement
To provide further context for those children who did move between quintiles across consecutive timepoints, the direction of movement was coded as either being positive (i.e., moving into a less deprived quintile) or negative (i.e., moving into a more deprived quintile) at each transition point. These proportions were calculated for all quintiles (see Fig. 5), as well as the middle three quintiles only (see Appendix C) < 0.001*** < 0.001*** < 0.001*** < 0.001*** < 0.001*** Note. Statistics calculated using the sample for analysis. Quintile 1 = least deprived neighbourhoods; Quintile 5 = most deprived neighbourhoods *p < .05. **p < .01. ***p < .001 to reduce the impact of potential floor and ceiling effects. The similarities between the two figures ( Fig. 5 and Appendix Fig. C1) suggest that the exclusion of the most extreme quintiles did not change the trend in the direction of movement.
Overall, and in alignment with Fig. 4, greater levels of movement were experienced in middle childhood compared to early childhood. The amount of upward and downward mobility experienced by children was roughly equal at each transition point, illustrated by the mirroring of the 'Up' and 'Down' datapoints. For example, at the final transition point 23.0% of children moved into a less deprived quintile while 25.9% of children moved into a more deprived quintile. The 'Average' trendline suggests very slight, but consistent, net movement into relatively more deprived neighbourhoods, compared to relatively less deprived neighbourhoods.
The micro-level analyses indicate that the majority of children did experience some level of movement between quintiles over the first 8 years of life but that the frequency of this movement differed depending on when this movement was experienced and what level of neighbourhood deprivation a child was moving from. Furthermore, roughly equal numbers of children were moving into more deprived neighbourhoods compared to less deprived neighbourhoods, supporting the conclusion drawn from the macro-level analyses that there was no clear pattern of upward or downward mobility experienced by this cohort of children.

Discussion
Tracking exposure to different residential contexts across the lifespan is an important component of neighbourhood effects research. More recently, sequence analysis has been used to aggregate cross-sectional measures of neighbourhood deprivation into a single variable that describes one's residential trajectory. In the current study, the findings of the sequence and cluster analysis indicate that children in Aotearoa NZ remained in similar levels of neighbourhood deprivation over time. The individuallevel analyses suggest that, despite more than three-quarters of children experiencing some level of neighbourhood mobility over the first 8 years of life, there was no obvious directional trend to this movement. While results of the macro-and micro-level analyses broadly align, the micro-level analyses provide context that is overlooked at the macro-level. This section firstly discusses the findings produced by each of the two methodological approaches individually, before exploring their collective contributions to the research field and our understandings of the main temporal patterns of exposure to neighbourhood deprivation experienced by children growing up in Aotearoa NZ. Lastly, the limitations and future directions are considered.

Investigating Exposure to Neighbourhood Deprivation at the macro-level
The sequence analysis findings posit that children experience minimal levels of neighbourhood mobility over the first 8 years of life. That is, there was no cluster characterised by a distinct pattern of movement into relatively more or less deprived neighbourhoods. While this finding does not align with similar studies of neighbourhood deprivation that utilise sequence analysis (e.g., , other studies of neighbourhood effects have come to similar conclusions. For example, Hagedoorn and Helbich (2021) measured exposure to residential contexts both longitudinally (averaged, cumulative, and weighted cumulative measures) and crosssectionally (current residential address) to determine whether neighbourhood effects related to health differed depending on the measure of residential context employed. Hagedoorn and Helbich (2021) found that cumulative measures of residential context were no more effective than the cross-sectional measure in predicting suicide mortality. Among their interpretations of this result was that the cross-sectional measure, while less detailed than the longitudinal measures, was an effective indicator of people's residential trajectories. Indeed, this was Pearson et al.'s (2013a) conclusion regarding their study of area-level deprivation and area-level mortality in which the model fit was better when current levels of deprivation were used to predict outcomes, rather than the longitudinal measure. Jivraj and colleagues (2019) suggest that the superior predictive capabilities of contemporaneous measures of neighbourhood deprivation are derived from earlier exposures that combine to form a 'chain of risk' (Kuh et al., 2003). Chain of risk models propose that exposure to one timepoint of deprivation increases the likelihood of experiencing another, resulting in a sequence of deprivation exposures, thus explaining the accuracy of cross-sectional measures of exposure in representing residential trajectories and in predicting relevant outcomes.
Furthermore, while the clusters identified in the current study do not match those presented in other studies of childhood exposure to neighbourhood deprivation, an important point of difference should be acknowledged: the current study analyses fewer timepoints of data (six) and over a smaller timeframe (0-8 years), compared to  and  who analysed data from Dutch governmental records collected annually from birth to 15 and 19 years respectively. Both of these studies identified clusters characterised by change, such as 'late deprivation' and 'late affluence', whereas the current study did not. The sequence index plots of these clusters suggested that the transition point at which the level of neighbourhood deprivation changed was during middle childhood: in  study, for example, children in the late deprivation cluster typically moved from middle-income neighbourhoods into deprived neighbourhoods at 7 years old, while children in the late affluence cluster typically moved from middleincome neighbourhoods into affluent neighbourhoods by 8 years of age. The current study similarly identified increasing rates of neighbourhood mobility in middle childhood, suggesting that the addition of neighbourhood deprivation data collected during the adolescent period may produce similar results to those produced by Kleinepier and colleagues (2018). Accordingly, more studies documenting childhood and adolescent exposure to neighbourhood deprivation using sequence and cluster analysis are required to further interpret and compare the current findings to other contexts.
In describing cluster membership, the current study found that experiences of neighbourhood deprivation were associated with mother sociodemographic characteristics. Children of Māori and Pacific mothers were overrepresented in more deprived clusters, especially in comparison to their NZ European counterparts. Kleinepier and colleagues (2018) similarly found that children from ethnic minority groups in the Netherlands were more likely to be exposed to greater levels of neighbourhood deprivation at any one time point and, in particular, were more likely to be consistently exposed to neighbourhoods with the greatest levels of deprivation. For example, after accounting for parental and household characteristics Turkish children were more than three times more likely to be in the consistent deprivation cluster compared to native Dutch children. One explanation for the residual effect of ethnicity in predicting cluster membership after controlling for family socioeconomic resources is that there are structural factors that influence where people live, such as racial discrimination . While arguably enacted through more covert and indirect means than historical discrimination, Pager and Shepherd (2008) found that racial discrimination continued to produce social and economic inequities in the areas of employment, housing, credit markets, and consumer interactions, in the U.S. Within the Aotearoa NZ context, as a settler-colonial state, racial discrimination related to homeownership has been used to marginalise Māori (Norris & Nandedkar, 2020). For example, Houkamau and Sibley (2015) found that self-identified Māori who felt that their appearance clearly reflected their Māori ancestry were, on average, less likely to own their own home compared to Māori who reported low levels of Māori prototypicality. Houkamau and Sibley (2015) summarised that the accumulation of past, and likely present, institutional racism had resulted in differences in rates of home ownership according to how Māori, or 'mortgage worthy', one appeared to be. Thus, there is an "inextricable relationship between whiteness, land/property, and power" (Norris & Nandedkar, 2020, p. 4) that underpins where one lives.
The study of neighbourhood effects has seemingly come at the expense of understanding people's residential decisions and "how people sort across geography" (DeLuca & Jang-Trettien, 2020, p. 451). Yet, Van Ham and Manley (2012) noted that studies that do not account for selection effects will bias the estimation of neighbourhood effects by failing to isolate the impact of the neighbourhood from factors that determine where families live in the first place. Slater (2013) went further and argued that the neighbourhood effects thesis should be completely inverted, shifting the assertion from 'the place one lives affects their life chances' to 'one's life chances affect where they live'. Indeed, the cumulation of the lack of opportunities for minority ethnic groups in the areas mentioned by Pager and Shepherd (2008), as well as in education, caused by racial discrimination can result in lower levels of wealth and thus a lack of residential mobility choice. To conflate such structural factors with the neighbourhood in which one lives will, therefore, misrepresent the impact of exposure to neighbourhood deprivation in childhood on health and development. Slater's (2013) approach to the study of neighbourhood effects emphasises the role of structural dynamics in shaping both people's residential experiences and their life outcomes, rather than stigmatising neighbourhoods and blaming the residents within them for the negative outcomes that they experience. This is particularly pertinent to interpreting children's experiences of neighbourhood deprivation given that while children do not likely participate in decisions about housing they are nonetheless impacted by these decisions.

Investigating Exposure to Neighbourhood Deprivation at the micro-level
The micro-level analyses of exposure to neighbourhood deprivation were conducted to explore timepoint-to-timepoint differences in residential contexts not captured by the sequence and cluster analysis findings. Levels of discontinuity between the six timepoints indicate that there were much higher levels of movement experienced in middle childhood, compared to early childhood. While the significantly higher level of quintile-quintile discontinuity between the 2-year and 54-month DCWs was expected compared to the earlier timepoints, the maintenance of this frequency was not. The use of the NZDep 2006 measure up until the 54-month DCW meant that the first two transition points captured exposure to different levels of neighbourhood deprivation caused by residential mobility only, while the differing measures of NZDep used at 2-years and 54-months captured quintile-quintile discontinuity caused by both residential mobility and change in neighbourhood context or composition. NZDep2013 was used for the final three timepoints. This suggests that the change in measure of neighbourhood deprivation may not have been the main driver of the increased levels of movement, but rather signals the start of the middle childhood period which is characterised by greater levels of neighbourhood mobility. This aligns with findings that show that children experience high levels of residential mobility (Morton et al., 2020;Rumbold et al., 2012). Indeed, children's commencement of formal schooling can motivate families to move into more desirable residential contexts for education (Hansen, 2014;Kim et al., 2005), as well as enable the primary caregiver to return to workforce, increasing the economic resources available to the family (Finseraas et al., 2017;Padilla-Romo & Cabrera-Hernández, 2019).
Additional micro-level analyses were used to contextualise the levels of quintilequintile discontinuity, finding that the likelihood of changing residential contexts depended on the context which they were moving from. That is, children from the middle three quintiles were more likely to move into a different level of neighbourhood deprivation compared to children in the most extreme quintiles. This may be due to floor and ceiling effects. Robertson and colleagues (2021) similarly found that children in Aotearoa NZ who had moved to a new house at least once over the first four years of life were most likely to move to a neighbourhood with the same level of deprivation as the one that they were moving from, particularly for those in the least and most deprived neighbourhoods. This was despite children from the most deprived neighbourhoods moving to a new house more often (i.e., greater levels of residential mobility) compared to children residing in less deprived neighbourhoods.
In the current study the observed floor and ceiling effects seemed to have an unequal impact on children's residential experiences: across the latter three transition points a greater proportion of children in Quintile 1 moved to a different quintile compared to children in Quintile 5, suggesting that it is harder to move out of areas with the greatest levels of deprivation than it is to move to an area of increased arealevel deprivation. Indeed, a study utilising earlier versions of the NZDep measure in Aotearoa NZ found that there was an inverse relationship between the decile of the neighbourhood a person was moving from and their chance of upward mobility, particularly for those in areas with the most deprived decile score (Morrison & Nissen, 2010). That is, it was less likely for individuals to move into relatively less disadvantaged neighbourhoods if they currently resided in more disadvantaged neighbourhoods. This floor effect is reflected in the overall net direction of movement being very slightly downward (i.e., into more deprived neighbourhoods), though this cannot be considered a trend given the near perfect mirroring of the proportions of upward and downward neighbourhood mobility over time. Accordingly, children growing up in Aotearoa NZ are more likely to be exposed to different levels of neighbourhood deprivation in middle childhood, but the level of deprivation children move to is equally likely to be higher or lower than their current residential context.

Comparing the macro-and micro-level Conclusions About Exposure to Neighbourhood Deprivation
While the macro-level findings lead to the conclusion that children in Aotearoa NZ experience little neighbourhood mobility over the first 8 years of life, the microlevel analyses suggest an increasing level of mobility over time. However, the movement in middle childhood was not characterised by a specific trend in direction, thus corroborating the absence of clusters that describe specific patterns of movement into lesser or greater levels of relative deprivation. When cluster membership was compared to children's cross-sectional exposure to neighbourhood deprivation the findings further echoed the sentiments of the micro-level findings: levels of quintilecluster discontinuity increased over time, reflecting the generally increasing levels of movement, and greater levels of discontinuity were observed among the middle three clusters compared to the two extremes, reflecting floor and ceiling effects. These results are antithetical to what would be expected given the regression toward the mean phenomenon, whereby individuals near the distributional extremes statistically show an increased likelihood of observations nearer the mean in subsequent observations (Nesselroade et al., 1980;Streiner, 2001). The observed 'stickiness' of being in the most extreme clusters (particularly the most deprived cluster), demonstrated both by the clusters identified and the lack of discontinuity between cluster membership and point-in-time deprivation, emphasises the challenges to upward mobility faced by those residing in the most deprived neighbourhoods. Lastly, when mother sociodemographic characteristics were used to describe cluster membership and quintile membership at each of the six timepoints, similar patterns of representation emerged. The current study suggests that macro-and micro-level analyses of residential trajectories provide complementary insights about the temporal patterns of exposure to various residential contexts experienced by children in early and middle childhood. While sequence analysis offers an effective approach to capturing people's residential trajectories, the micro-level findings provide some important and more nuanced descriptions of the lived experiences of children in Aotearoa NZ that add depth to the interpretation of the identified clusters. Accordingly, the inclusion of such descriptive statistics to future studies that utilise sequence analysis may prove a useful addition.

Limitations and Future Directions
The current study is not without its limitations, many of which are derived from utilising a secondary dataset. Firstly, the current study did not determine why children and their families did or did not move into different residential contexts, which is required to develop a theory of selection bias (van Ham & Manley, 2012). The GUiNZ study has collected data related to this topic, though not for every DCW. Developing a greater understanding of selection effects and incorporating such effects into appro-priate statistical models to estimate neighbourhood effects remains a fundamental challenge for researchers to address, thus providing an ongoing opportunity for future research in general and the GUiNZ dataset in particular.
Secondly, the change in NZDep measure between the 2-year and 54-month DCWs introduced an additional source of variation which made this transition point difficult to interpret in terms of the relative contribution of residential mobility and change in neighbourhood context or composition in causing children to be exposed to different levels of neighbourhood deprivation, compared to the other transition points that utilised the same NZDep measure (thus isolating the impact of residential mobility). However, the findings suggest that the impact of this change in measure was small.
Thirdly, the authors wish to draw the reader's attention to the terminology used in neighbourhood effects research and the interpretation of such terms within the current study. 'Neighbourhood deprivation' is a common term in this line of research likely due to the large proportion of studies investigating the impact of current and cumulative exposure to greater levels of neighbourhood deprivation in childhood on health and wellbeing, in an effort to identify causal mechanisms that can be augmented to improve outcomes. The term, though, is a vague one, and may be interpreted incorrectly if not specifically defined. In the current study the measure of neighbourhood deprivation used captures the degree to which a neighbourhood is lacking in a specific subset of socioeconomic resources. This does not mean that residential contexts with high deprivation scores lack in other social, economic, and cultural assets which promote a high quality of life. Furthermore, while neighbourhood effects research generally find that living in areas with greater levels of deprivation increased the likelihood of negative outcomes for residents, it is important to note that this is not deterministic. Indeed, Pearson and colleagues (2013b) found many examples of neighbourhoods that experienced low levels of mortality despite their high levels of deprivation.
Lastly, the measure of neighbourhood deprivation utilised in the current study is a relative one: irrespective of whether neighbourhoods in Aotearoa NZ have become less deprived in absolute terms there will still be 20% of neighbourhoods defined as being in the most deprived quintile. However, there is still value in tracking changes in exposure to neighbourhood deprivation over time utilising a relative measure because it provides a description of neighbourhood mobility that is relevant to the local context, such as the identification of different patterns of mobility and how these patterns are experienced by different groups of people, in this case children.

Conclusion
The study of neighbourhood effects is complex. It requires an understanding of neighbourhood characteristics, the temporal patterns of exposure to residential contexts, and the demographics of the individual to account for the heterogeneity of neighbourhood effects, identify causal mechanisms, and develop a theory of selection bias. The recent application of sequence analysis to neighbourhood effects research has provided an efficient way to collate multiple timepoints of data into a single variable that captures one's residential trajectory. Exploring the main temporal patterns of exposure to neighbourhood deprivation at the macro-level, however, may come at the expense of more granular observations related to the way in which people move in and out of different residential contexts over time. Accordingly, the current study sought to investigate the ways in which children in Aotearoa NZ were exposed to different residential contexts over the first 8 years of life using two different methodological approaches: by analysing longitudinal data at the macro-level, using sequence and cluster analysis, and at the micro-level, by focussing on the nature of individual timepoint-to-timepoint movement between levels of neighbourhood deprivation. The study also sought to compare the findings produced by each level of analysis.
The sequence and cluster analysis identified five clusters which generally aligned with the five levels of neighbourhood deprivation measured, suggesting that at the population-level, children experience little neighbourhood mobility over time. At the micro-level, the findings demonstrated an increase in levels of movement between different levels of neighbourhood deprivation in middle childhood. However, this movement lacked a specific directional trend. The micro-level analyses also found that the likelihood of moving to a different residential context depended on the level of deprivation one was moving from, making it harder for children living in the most deprived neighbourhoods to move into relatively less deprived neighbourhoods. Overall, the current study found that the macro-and micro-level findings produced complementary insights about children's residential experiences in Aotearoa NZ and encourages future studies of neighbourhood effects to include findings from both levels of analysis in order to provide a more comprehensive description of children's residential trajectories and their outcomes.
is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/ licenses/by/4.0/.