Common Descriptive Statistics
Except for Chapter 4, which only uses the New Zealand General Social Survey (NZGSS) dataset, all of the chapters in this thesis utilise the NZGSS matched with the 2006 Census. Thus, many of the variables in these chapters are the same, especially those derived from the NZGSS. To avoid repeating the common sampling issues, the common context is discussed in the current chapter. The data collection process of the NZGSS dataset is explained in Section 3.1.
Due to computational limitations – given the complicated spatial analysis in Chapters 7, 8 and 9 – the sample is limited to Auckland city. To maintain consistency, the sample used in Chapter 6 is also limited to Auckland. However, to give a sense of the differences between the results derived from the entire sample and the Auckland sample, the results of Chapter 6 are also derived based on the New Zealand sample and presented in the appendix to Chapter 6 (see Appendix i). The common descriptive statistics based on the Auckland sample are presented in Section 3.1.1.
The first dataset I will use is the New Zealand General Social Survey (NZGSS) by Statistics New Zealand. I have used the full file of the NZGSS, i.e. the CURFs provided by SNZ do not contain all the variables that I have used in the current study. The NZGSS was carried out in three series, in April 2008, October 2010 and April 2012.20 The sample size for each cross-section was approximately 8500, a number that Statistics New Zealand (SNZ) selected to represent the total population. The data were collected by a face-to-face interview in which the interviewer used a laptop. Each household was interviewed once and the average interview time per household was 45 minutes. The survey consists of two separate parts: the household questionnaire (3.1.1) and the personal questionnaire (3.1.2).
The other dataset I used is the New Zealand Census 2006 (also conducted by SNZ), which is an official count of the population and dwellings in New Zealand. This dataset contains data at the mesh-block level, which is the smallest geographical level at which the data is processed by SNZ. I matched the NZGSS with this dataset in order to determine the urban area in which the individuals are located. I explain the matching process in Section 7.4.
The household questionnaire
The first questions of the survey are about members of the household and include sex, age and ethnicity. Based on some eligibility rules, the household members, who are in scope and are available during the interview period, answer these questions. The computer selects one of the members to answer the next part, which is the personal questionnaire.
The personal questionnaire
This questionnaire consists of 14 different topic modules. In this study, in addition to the core personal data, satisfaction with living environment, health, skills and abilities are considered. Residential satisfaction was described earlier in Section 2.2. Health perceptions are derived from a question that asks, ‘In general, would you say your health is excellent, very good, good, fair or poor?’ and skills perceptions are derived from a question that asks, ‘How do you feel about your knowledge, skills and abilities?’ Respondents are questioned about their feelings regarding their living environment as at the time of the interview and can choose from among five responses ranging from ‘very satisfied’ to ‘very dissatisfied’. The core personal data are mostly categorical. For example, household income is measured based on the New Zealand Standard Classification of Income Bands 2009, which has 16 categories.
Resampling by using replicate weights
Different methods of resampling may serve to increase the precision of estimation using survey designs by deriving more robust standard errors, proportions, odds ratios and regression coefficients. To do this, a random set of observations is left out at each time of estimation. Replicating this process leads to the estimation of the bias of a statistic; this method is called jackknifing. The NZGSS provides us replicate weights produced by the delete-a-group jackknife method (Kott, 2001). In the dataset, 100 groups are derived by using primary sampling units (PSUs) randomly sorted into each stratum. This strategy results in 100 replicate samples in each, of which one of the groups is omitted and weights are adjusted accordingly. Using these weights in our estimation leads to estimates that tend asymptotically to true values.
The geographic scales
In order to have a reliable definition of the neighbourhood, we should rely on the resident’s perception of their neighbourhood boundaries that can be derived from face-to-face interviews (Witten, McCreanor, & Kearns, 2003) 7.4.2. A typical definition of neighbourhood consists of a certain number of nearest neighbours. This definition is very similar to the mesh-block’s definition as an area consisting of dwellings that are located close to each other. Mesh-block is defined by SNZ as a very small neighbourhood, which consists of a few houses located next to each other, with an average population of 50 people.21 Because I assume that the reference group consists of people who live shoulder-to-shoulder with an individual, the mesh-block design suits our purpose well.
In the context of New Zealand, larger neighbourhoods are defined as area units, which approximate city suburbs (as opposed to administrative boundaries). Most people consider themselves as members of their suburb or area unit. Hence, to take perceptions into account, area units may also be considered as a good definition of neighbourhoods. For illustrative purposes, the area units located in central areas of Auckland are depicted on the left-hand side of figure 8. On the right-hand side, the mesh-blocks located in the same area are depicted. As will be discussed in Sections 7.2.1 and 5.4.2, however, individuals’ perceptions of their neighbourhoods’ boundaries depend heavily on their characteristics, such as their education and income levels. On a larger scale, I consider the boundaries of territorial authorities (TLAs) – that is, local councils
which are administratively determined community of interest that share a common local government. In other words, people with similar ideas and thoughts about a certain topic are considered to locate in the same territory. Similar people may tend to congregate in particular neighbourhoods, and it is an empirical matter as to which neighbourhood definition is most appropriate in this context.
The datasets used in the current study are three cross-sections of the New Zealand General Social Survey from 2008 to 2012 and Census 2006, both provided by Statistics New Zealand. This provides our study with 8500 observations. In order to derive robust results, I will use the replicate weights produced by the delete-a-group jackknife method. In order to find the correct definition of neighbourhood boundaries, I consider three conventional definitions of boundaries at the mesh-block level, the area-unit level and the territorial authority level.
Mesh-blocks are very small neighbourhoods that consist of a few houses located next to each other, with an average population of 50 people. Area units are aggregates of mesh-blocks and are the suburbs that people usually consider themselves members of. Territorial authorities are very large areas determining communities of interest that share a common local government.
All variables in Table 2, descriptive statistics, are categorical variables. For example, ‘Gender’ is a binary variable with 0 for females and 1 for males (in NZGSS, this variable is called ‘cordv10’, as shown in the brackets). The mean of this variable is equal to 0.485, indicating that 48.5 per cent of respondents are male and the rest are female. On the right-most column, the equality of the mean of each variable between three cross-sections of data; – 2008, 2010 and 2012 – is tested. The significance level of less than 0.001 for the ‘Age’ variable, which is an indicator of respondents’ age, statistically violates the null hypothesis that there is no change in the mean of this variable across the three surveys; in other words, it indicates that the percentage of people distributed in different age groups changes amongst these three cross-sections of data.
To avoid a very long list of variables in the regression tables, some of the variables with similar effects have been recategorised. In the ‘Group’ column of table 2, the group of each category is shown amongst different variables. For example, for ‘Homeownership status’, group ‘i’ includes categories ‘Owned, not defined’, ‘Owned, mortgage’ and ‘Owned, no mortgage’, and so on.22 Two criteria are used for grouping some categories. The first is the diversity of the category; for example, MELAA ethnicity, which consists of Middle Eastern, Latin American and African ethnicities, is a highly diverse category and is therefore grouped with ‘Other’. The second criterion is the results derived from different categories; two categories are grouped if they show very similar patterns in regression models.
In a comparison between three cross-sections, for almost all residential issues, a lower fraction of respondents report problems with their neighbourhoods in 2010. For most housing problems, however, this difference is not statistically significant at the 1 per cent level. Three variables in this table are based on a five-level Likert scale: health perception, ability perception and residential satisfaction. Health and ability perception variables are of interest here to address the potential ‘over-optimism’ bias, as is discussed in the next section.
1.2 Motivation and contribution
1.3 Thesis outline
2 Common Literature
2.1 Auckland, the City of Sails
2.2 Residential satisfaction measurement
2.3 Factors affecting residential satisfaction
2.4 Neighbourhood effects
3 Common Descriptive Statistics
3.1 Data sources
3.2 The geographic scales
3.3 Descriptive statistics
4 The Outcome Of Interest
4.1 Hedonic demand theory
4.2 Subjective measures and the economics literature
4.3 Features of subjective measures
4.4 Are subjective measures superior to objective ones?
5 Common Methodology
5.1 Utility function
5.2 The econometric basis for welfare maximisation
5.3 Changing the ordinal dependent variable to a binary variable
5.4 Addressing the potential issues of subjective measures
6 Is Residential Satisfaction Affected by Crowding? A Study of Raw, Adjusted and Perceived Crowding Measures
6.2 Literature review
6.3 Hypothesis development
6.4 Data and sample
6.5 Descriptive statistics
6.6 Research design
6.7 Models and results
7 Residential Satisfaction, Household Crowding and Density: Evidence over Different Geographic Scales in Auckland
7.2 Literature review
7.3 Hypothesis development
7.4 Data and sample
7.5 Descriptive statistics
7.6 Research design 24
7.7 Models and results
8 Residential Satisfaction, Relative Crowding and Relative Density: Evidence over Different
8.2 Hypothesis development
8.3 Data and sample
8.4 Descriptive statistics
8.5 Research design
8.6 Models and results
9 Negative Envy or Positive Amenity Effects? A Neighbourhood Study of Aucklanders’ Residential Satisfaction
9.2 Literature review
9.3 Hypothesis development
9.4 Data and sample
9.5 Descriptive statistics
9.6 Research design
9.7 Models and results
10 SPWT: A Stata Command for Creating Spatial Weight Matrices
10.2 SPWT’s weighting methods
10.3 help: spwt
11.2 Summary and discussion
11.3 Limitations of the current study
11.4 Suggestions to the data provider
11.5 Policy implications
GET THE COMPLETE PROJECT
EFFECTS OF CROWDING, DENSITY AND DEPRIVATION ON RESIDENTIAL SATISFACTION