Official Statistics Research Series, Volume 5, 2009

Contents

1. Does Mortality Vary Between Pacific Groups? Estimating Samoan, Cook Island Maori, Tongan and Niuean mortality rates using hierarchical Bayesian modelling.
2. Confidentialising Microdata Using Multiple Imputation: Development and evaluation of a non-parametric hierarchical Bayesian imputation model for numerical data.
3. Generating synthetic microdata from published marginal tables and confidentialised files
4. Geospatial Land-Use Classification for New Zealand: Review and Recommendations
5. Sustainable Development and Equity
6. Sustainable Development and Cultural Capital

1. Does Mortality Vary Between Pacific Groups? Estimating Samoan, Cook Island Maori, Tongan and Niuean mortality rates using hierarchical Bayesian modelling (PDF, 608KB)
Tony Blakely, Ken Richardson, Jim Young, Patrick Graham, June Atkinson, Paul Callister, Robert Didham, & Martin Tobias

Abstract: Pacific mortality rates are traditionally presented for all Pacific people combined, yet there is likely heterogeneity between separate Pacific groups. We aimed to determine mortality rates for Samoan, Cook Island Maori, Tongan and Niuean people. We also aimed to test and demonstrate the application of hierarchical Bayesian methods to sparse data. We used New Zealand Census-Mortality Study (NZCMS) data for 2001-04, for 380,000 person years of follow-up of 0-74 year olds in the 2001-04 cohort for which there was complete data on sex, age, ethnicity (total counts), country of birth and household income. Given sparse data, we used hierarchical Bayesian (HB) regression modelling, with: a prior covariate structure specified for sex, age, country of birth (CoB) and household income; and smoothing of rates using shrinkage. The posterior mortality rate estimates were then directly standardised. Standardising for sex, age, income and CoB, all-cause mortality rate ratios compared to Samoan were: 1.21 (95% credibility interval 1.05 to 1.42) for Cook Island Maori; 0.93 (0.77 to 1.10) for Tongan; and 1.07 (0.88 to 1.29) for Niuean. Cardiovascular disease (CVD) mortality rate ratios showed greater heterogeneity: 1.66 (1.26 to 2.13) for Cook Island Maori; 1.11 (0.72 to 1.58) for Niuean; and 0.86 (0.58 to 1.20) for Tongan. Results were little different standardising for just sex and age. We conducted a range of sensitivity analyses about a plausible range of (differential) return migration by Pacific people when terminally ill, and a plausible range of census undercounting of Pacific people. Our findings, in particular the elevated CVD mortality among Cook Island Maori, appeared robust. To our knowledge, this project is the first time in New Zealand that clear (and marked in the case of CVD) differences in mortality have been demonstrated between separate Pacific ethnic groups. Future health research and policy should, wherever possible and practicable, evaluate and incorporate heterogeneity of health status among Pacific people. 

Keywords: Pacific Islanders, mortality rates, hierarchical Bayesian modelling.

2.Confidentialising Microdata Using Multiple Imputation: Development and evaluation of a non-parametric hierarchical Bayesian imputation model for numerical data (PDF, 776KB)
Patrick Graham, Lisa Henley, & Lena Rodnyanskiy

Abstract: When developing data products for use by external researchers, statistical agencies seek to meet legal and ethical commitments to respect respondent confidentiality by modifying the data originally collected. Statistical agencies also need to ensure that data released to researchers can be relied upon to produce valid inferences across a broad range of statistical analyses. To some extent these aims are in competition because modifying data before release may distort some inferences obtained from the released data. Multiply imputed synthetic data has been proposed as a solution to the data release problem faced by statistical agencies. In this report we summarise the background to the multiple imputation proposal and develop new multiple imputation methods for producing synthetic versions of data comprising a mix of categorical and numerical variables. The imputation models developed build on an earlier project which proposed hierarchical Bayesian imputation models for categorical data. This framework is extended to data comprising both categorical and numerical variables by modelling the joint distribution of numerical variables conditionally on categorical variables. Hierarchical Bayesian models for numerical variables are extended by using ideas and models from non-parametric Bayesian statistics. The resulting imputation method uses a generalised Polya urn sampling scheme which creates synthetic data as a mix of model-generated records and records sampled from the data. The performance of the proposed new imputation methodology is evaluated in two sets of comparisons. In the first, the new imputation method is compared with fully parametric hierarchical model based imputation in an exercise which uses the CURF for the 2003 Household Income Survey, as the real data to be confidentialised. In the second, the original 2003 Household Income Survey data is used as the real data and the performance of multiply imputed synthetic versions of this dataset, the CURF for this dataset and another confidentialising method, known as sufficiency based perturbation are compared.

Keywords: Confidentiality, hierarchical Bayes, multiple imputation, non-parametric Bayes, synthetic data.

3. Generating synthetic microdata from published marginal tables and confidentialised files (PDF, 523KB)
Alan Lee

Abstract: We describe several methods for generating synthetic data sets, using a combination of publically available marginal tables, and microdata samples. Our methods are based on fitting parsimonious statistical models to high-dimensional tables of relative frequencies, and then generating synthetic data from these models. We describe a set of R functions which implement the methods under study, and apply the methods to data from the 2001 Census of Population and Dwellings.

Keywords: Contingency tables, marginal tables, log-linear models, mixture models, maximum likelihood, iterated proportional fitting, Fisher scoring, latent class analysis, Metropolis-Hastings algorithm.

4. Geospatial Land-Use Classification for New Zealand: Review and Recommendations (PDF, 648KB)
Daniel Rutledge, Robbie Price, Sarah Cowell, & Craig Briggs

Abstract: This project examined theoretical and practical approaches to land-use classification to support development of a geospatial land-use classification for New Zealand. Land use refers to the modification of and related activities on the land by people to sustain human life. We allocate and change land use to produce different combinations of goods and services in response to changing human needs and wants. Some changes are temporary, e.g., crop rotation, while others are more permanent, e.g., urbanisation. Also, some consequences of land-use change or associated practices are easily observed, e.g., forest harvesting, while others are not, e.g., gradual loss of soil fertility through inappropriate management. Given that New Zealand currently lacks nationally consistent and comprehensive geospatial land-use information, developing a geospatial land-use classification would help fill that gap by 1) meeting a range of needs, 2) helping underpin policy, planning, and resource management, and 3) contributing to reporting on progress towards sustainable development across scales within New Zealand.

Land-use classification broadly falls into four types along a gradient from simple to complex. Categorical approaches depict land use as discrete classes with no relationships among them. Hierarchical approaches depict land use as a nested set of classes. Higher level classes, e.g., urban, become divided into more specific classes, e.g., residential, commercial. Multidimensional approaches store multi-attribute information for defined unit areas and allow recombination to generate different land use classes. Semantic classifications derive from linguistic theory and describe land use using a rich collection of words and formal grammar from which different interpretations (i.e. classifications) could be derived. International practice strongly favours the use of hierarchical classifications, and many countries have adopted official hierarchical land use classifications. A few examples (e.g., SIOSE from Spain, American Planning Association) take a more multidimensional approach. Regardless of the approach, most land-use classifications internationally remain unimplemented or partially implemented. In New Zealand, there have been four attempts to develop a land-use classification. None of these became officially adopted, although current land valuation practices include an unofficial land use classification system. Our results suggest New Zealand would benefit from adopting a semantic approach to land-use classification. Such an approach appears the most promising, given its ability to include a diverse range of information, to incorporate new information as it becomes available, and to generate a range of classifications, including reference or official classifications, to meet a variety of needs while retaining the capacity to translate and compare among them.

Keywords: Categorical, classification, geospatial, hierarchical, land cover, land use, land-use activity, land-use change, semantic, multidimensional, New Zealand.

5. Sustainable Development and Equity (PDF, 520KB)
Paul Dalziel, Caroline Saunders, Rosie Fyfe, & Bronwyn Newton

Abstract: This report draws on the New Zealand experience to explore how intra-generational equity might relate to sustainable development in a New Zealand context. After a summary of different approaches to equity found in the international literature, the report considers Article 3 of the Treaty of Waitangi, which imparted to Māori all the rights and privileges of British subjects. The report then discusses how this Article’s widely accepted interpretation as implying that the Crown must equally protect the social rights of all New Zealand citizens, has been applied in the creation of a system of social security to protect New Zealand citizens against poverty and social exclusion. This discussion focuses on the central roles of employment, income support, housing, health and education. This is followed by a survey of a recent debate in New Zealand that focused specifically on pay and employment equity.

This approach is adopted as the basis for the proposals towards an equity indicator made in the report. It takes the core elements of New Zealand’s social security – employment; income; housing; health and education – and defines standards for what would be meant by equity for each element. The chapter proposes a statistical series for measuring departures from each standard and then combines these statistics into a proposed equity index. The index is illustrated using New Zealand data in 1996, 2001 and 2006. Data for 2006 record almost no disparity between the male and female populations, but a substantial difference between the European/Pākehā and Māori populations.

Keywords: Equity index, sustainable development, well-being.

6. Sustainable Development and Cultural Capital (PDF, 589KB)
Paul Dalziel, Caroline Saunders, Rosie Fyfe, & Bronwyn Newton

Abstract: It has become common for countries to create a framework of sustainable development indicators (SDI) to measure progress in raising the social, economic and environmental well-being of their regional and national communities. Standard practice is to focus on changes in stock measures of physical capital, financial capital, human capital, natural capital and social capital. An unusual feature of New Zealand’s approach is that it pays explicit attention to cultural well-being alongside social, economic and environmental well being. This practice raises important issues about how measures related to cultural well-being can be incorporated into a national SDI framework. The research for this report explored whether cultural capital is best conceptualised as a component of social capital, or whether it warrants treatment as a separate category. In arguing for the latter the report proposes that cultural capital be defined as a community’s embodied cultural skills and values, in all their community-defined forms, inherited from the community’s previous generation, undergoing adaptation and extension by current members of the community, and desired by the community to be passed on to its next generation.

Keywords: Bourdieu, cultural capital, indicators, indigenous, sustainable development, well-being.