Sally Casswell and Krishna Bhatta
Alcohol & Public Health Research Unit, May 2001


The sampling frame was the free-dialling area of the Auckland metropolitan region.  Interviewing was undertaken using the Alcohol & Public Health Research Unitís in-house computer-assisted telephone interviewing (CATI) system.  This is a network of 20 computer stations and a supervisorís station.  The questions were programmed into the computer and the interviewers read them as they appeared on the screen. They then coded the answers directly into the computer, from the options presented on the screen.  The supervisors were at any time able to observe any interview on their own screen and listen in to any call without the interviewer or respondent being aware (respondents were told this might happen before they began).  The ability to monitor all interviewing at the one central location ensured a high degree of quality control.  Quality was also emphasised in the extensive training given to interviewers.  To ensure further quality of the survey data, the information collected was checked by the CATI supervisors to identify invalid, missing or any dubious figures.  If required, call backs were arranged to check the recorded information.
Telephone numbers were randomly selected. At least ten calls were made to each household to try and obtain an interview with the qualifying respondents.  These calls were made at different times of the day and during both week days at the weekends.  The random digit dialling system ensured that unlisted numbers, estimated at 10% of domestic phones, were included in the sampling frame.

  When a household was contacted, initials of all members aged 14-65 years were collected to allow the computer to randomly select the interviewee.  Any number of the eligible residents could be selected by the computer, and in some cases it selected none.  On average every 2.5 persons in the eligible age range was selected. People were told that the survey was being undertaken by the Alcohol & Public Health Research Unit from the University of Auckland Medical School.  They were also told that they could be interviewed by a Maori interviewer if they wished.  Confidentiality was assured and only first names were collected.

Statistics New Zealandís Household Economic Surveys show that telephone penetration in New Zealand is comparatively high.  Over the period 1990 to 1996 it ranged from 93.5% to 96%.  Certain sectors of the population, most notably Maori and people from the Pacific Islands, are under-represented among telephone owners, and they are the same groups who tend to be under-represented in face-to-face surveys.  Analyses undertaken by the Alcohol & Public Health Research Unit (Wyllie et al., 1994) have shown that this under-representation has little effect on results at a population level.



In an effort to determine whether the data from the annual Auckland alcohol surveys is a valid representation of patterns of change, comparisons were made with other data relating to alcohol consumption and purchasing.  Although direct comparisons cannot be made, they provide useful indications.

Sensitivity of Alcohol Consumption Measures

The choice of questions for measuring alcohol consumption in the Auckland surveys was made after careful consideration of the advantages and disadvantages of various approaches (Wyllie et al., 1994).  The questions used are based on typical drinking behaviour at the different locations and it is possible that they are not sufficiently sensitive to detect small changes in quantities being consumed.  For example, a person could have more than one drinking pattern at home and change the quantities they consume in one of them, but if this is not their most typical pattern at home the change will not be detected by the survey questions.  Likewise, as a result of compulsory breath testing, a person may sometimes not drink at locations away from home, but their typical quantity at those locations may remain unchanged.
However, other measures in the surveys show patterns that are consistent with the consumption measures, which suggests that their sensitivity is not a major issue (see Internal Validity below for further discussion of this issue).

Possible Sampling and Measurement Errors

Changes in data can result from changes in research methods, questionnaire design, interviewing procedures, or variations in the samples.  Considerable efforts have been made to minimise any change due to these factors. 

 Interviewer quality control

As noted above, there is considerable emphasis placed on interviewer quality control.  As well as the monitoring of work, this included each new interviewer receiving over 20 hours of training and experienced interviewers having time to familiarise themselves with the questionnaire prior to each survey. 

Questionnaire design and administration

There have been no changes to the questionnaire which would have affected the consumption questions.  There was a change in the computer software in 1995 and a change in the way in which random numbers were generated, but neither of these changes was expected to have affected the data obtained. 

Survey timing

As mentioned in the Research Methods section , there was found to be variation in the consumption data by time of interview.  This has been controlled for in the analyses.

Response rates

Variation in response rates could affect the data if those persons with higher response rates are different from those with lower response rates.  As shown in table D1, the response rate was lowest in the first year and was also lower than usual in 1993, 1997 and 1998.  However, none of these years have produced unexpected data.

Table D1 


Sample size

Response rate































Sample composition

A comparison of the composition of the samples in each survey shows that between 1994 and 1999 the proportion of the sample in the $40,000 and over income category has been increasing.  Since 1990, the proportion has been: 19%, 18%, 17%, 18%, 21%, 22%, 24% 25%, 27% and 31%.  To check this trend against other data sources, data were purchased from the Household Economic Survey (HES) undertaken by Statistics New Zealand  (after 1998, HES figures went to 3 yearly, so 1999 figures were not available for this report.) 
Figure D1 shows the proportion of respondents reporting personal incomes of $40,000 and over in the Auckland surveys, compared with similar data for Auckland and New Zealand from the nationwide HES.  Auckland alcohol survey proportions were higher because the HES included persons aged over 65 years.  The trend in the Auckland alcohol data was broadly parallel to the trend in the HES data.  There were minor differences in the years 1991, 1993 and 1997, but they were very small, suggesting that the Auckland survey was sampling income groups adequately.  

Internal Validity

Data can be validated not only by comparisons with external criteria, but also by examining whether there is internal consistency in the data that is collected.  If there was decreasing alcohol consumption in Auckland over time, one might have expected to see some upward trends in the reasons for drinking less and downward trends for some of the reasons for drinking more.  There were in fact no such trends; rather there were trends in the opposite direction in keeping with the self reported consumption in the survey.  Also, the level of problems showed increasing rates, which is not what would be expected if consumption was actually decreasing. 
The 1995 annual Auckland survey data was collected as part of the 1995 National Drinking Survey.  In this survey the proportion of Aucklanders reporting drinking more was higher than for the rest of the country.  This is consistent with the Auckland survey data over those years, which did not show the downward trend evident in the national available alcohol figures.



Three major statistical analyses were performed in the tracking trend data: analysis of covariance (ANCOVA) to compare the group means; logistic regression analysis to compare the group percentages; and general linear regression model (GLM) to test a linear trend. All analyses were controlling for gender, age and interview date.
The assumptions of parametric ANCOVA are that 1) each group are normally distributed; 2) all groups have equal variances; 3) all covariate variables have a linear relationship with the dependent variable; 4) all groups' slopes are equal for each covariate variable.
When undertaking ANCOVA, all continuous variables needed to be transformed to make the data more nearly normal and reduce the variance. Logarithmic transformation was used for this purpose.  Because of the number of comparisons being undertaken, Bonferroni significance levels have been used, which adjust for the number of comparisons.
A major issue with data on drinking patterns is how to report it in a way that is both easy for people without statistical skills to comprehend, but also allows statistical significance testing and presents meaningful data.  A problem with using arithmetic means (averages) is that extreme values have a large influence. A few extreme values in a sample can have an influence on the data for the whole sample which would not accurately reflect the sample as a whole.  The alternative that was used in the reporting of the 1995 national survey Drinking in New Zealand was to use medians, the point at which half the sample is above and half below.  While medians are useful for reporting one-off studies, they have limitations for comparing surveys across time. For example if heavier drinkers were increasing their consumption more than lighter drinkers, this might not be reflected in a change in the median and yet overall consumption could show a large increase.  The current study has utilised means that have been subjected to logarithmic transformation; these means will be referred to as geometric means.  The logarithmic transformation reduces the influence of the more extreme values. 


Data was collected for drinking in each of fifteen standard locations plus any additional drinking locations identified by the respondents. 
The average Quantity Consumed on a Typical Occasion for an individual is the weighted average of the quantities consumed on a typical occasion at each location, taking into account how often the person drank at the location (frequency of drinking).  In this way, a location that a person only drank at once a year had minimal influence, compared with a location that the person drank at daily. 
The volume of absolute alcohol consumed on a typical occasion was then calculated from the types of drinks and volume of beverage drunk.  Beverage volumes were calculated from numbers and types of containers.  The volume of absolute alcohol in each type of beverage was calculated using Statistics NZ conversions.
The annual volume of absolute alcohol consumed in each location by an individual  was calculated by taking sum of the volumes of absolute alcohol consumed per occasion for each location, multiplied by the frequency of drinking at that location. The sum of all location amounts gives the annual volume of total absolute alcohol consumed by an individual.
The frequency of drinking was the sum of the frequencies at all of the locations.

   Top | Back | Home