New Zealand Gambling Survey

Overview of analysis in Kupe

Ethnicity

Both total response and prioritised ethnicity have been used in Kupe:

• Prioritised ethnicity means that a respondent is allocated to a single ethnicity even when they reported belonging to more than one ethnic group. The prioritised order is Māori, Pacific peoples, Asian peoples, then NZ European/MELAA/Other.
• Total response ethnicity means that people who reported belonging to more than one ethnic group are counted once in each group they reported.
• Equity-based ethnicity is another form of prioritised ethnicity. We use this for comparison purposes, and you will only see this in the 'Subgroups comparison' tab. For Māori equity, respondents were allocated in the priority of Māori, Pacific peoples, and then everyone else (non-Māori non-Pacific). For Pacific peoples, respondents were allocated in the priority of Pacific peoples, Māori, and then everyone else. For each equity variable, non-Māori non-Pacific is the comparison group, ie Māori versus non-Māori non-Pacific

Further details of these output options are in the Health Information Standards Organisation (HISO) Ethnicity Data Protocols (Ministry of Health, 2017).

Deprivation

NZDep is a small-area-based index. It provides a measure of neighbourhood deprivation. This is done by looking at the comparative socioeconomic positions of small areas and assigning them decile numbers, from least deprived (1) to most deprived (10). The index is based on nine socioeconomic variables from the Census. Deciles were grouped for analysis into least deprived (deciles 1-3) mid deprived (4-7) and most deprived (8-10). Click here for more information.

Subgroups comparison for binary indicators

These analyses aimed to understand how the risk by each indicator varied across demographic subgroups (such as gender) while adjusting for other factors such as age. We used a quasi-Poisson regression model with a logarithm link function (Lumley, 2011) to estimate relative risks (RRs) and related 95% confidence intervals (CIs) for binary indicators. We replaced estimates with dashes when we had any of the following indications of unreliability:

• The sample size of fewer than 10 events of interest per parameter in the model. This follows the guideline for logistic regression outlined in Hosmer et al. (2013), pp 407-408.
• The CIs that were unusually wide, which was defined as the width of the CI is greater than or equal to 4 times the ratio estimate.
• We applied a conservative approach to ratio estimates and suppressed those that were greater than 7 or less than 1/7.

Changes over time

The proportion or prevalence of each indicator was compared with the most recent available survey year if at least two data points were available. We suppressed all values that contain a small sample size ( n< 30) to maintain the reliability of results and minimise the margin of error. Some of the variation of estimated proportion or prevalence between survey years could potentially be from changes in the questionnaire and/or methodology.

Weighting adjustment

Statistical selection weighting adjustments were applied to each dataset to compensate for selection bias. Post-stratification weight was used to ensure that findings from the survey are representative of the New Zealand population with respect to major demographic characteristics such as gender, age, and ethnicity.

Reliability of survey results - sampling error

We present 95% CIs to indicate the uncertainty in an estimate due to collecting data from only a sample of the population. For survey data, 95% CI gives the range that if we select 100 different samples, we would expect the proportion value to fall within the range 95% of the time.

Reliability of survey results - non sampling error

Findings are likely to under- or overestimate some indicators due to the nature of self-reported information. For instance, when a question was asked 'thinking about the last 12 months, how often have you felt that you might have a problem with gambling?' the respondents then responded with either never, sometimes, most of the time, or almost always. Depending on what the respondent considers to be socially desirable, this can lead to over-reporting of good behaviours or under-reporting of risk behaviours. Also asking for the last 12 months assumes respondents can accurately recall previous events over the time period which may not be the case.

Survey results show associations rather than cause-and-effect relationships

Kupe provides a snapshot of data at one point in time. Results can be used to look at associations between different factors, such as alcohol consumption and neighbourhood deprivation. It does not look at cause-and-effect relationships.

For example, if we find out that a alcohol consumption is more common in people living in deprived areas, it does not mean that alcohol consumption is caused by living in deprived areas.

References

Lumley, T. (2011). Complex surveys: a guide to analysis using R(Vol. 565). Hoboken, New Jersey: John Wiley & Sons.

Ministry of Health. (2017).HISO 10001:2017 Ethnicity data protocols. Wellington: Ministry of Health. www.health.govt.nz/publication/hiso-100012017-ethnicity-data-protocols

Hosmer, D. W., Lemeshow, S., & Sturdivant, R. X. (2013). Applied logistic regression (Third edition). Hoboken, New Jersey: Wiley.

Korn, E. L., & Graubard, B. I. (1998). Confidence intervals for proportions with small expected number of positive counts estimated from survey data. Survey Methodology 24(2): 193-201.