Notes compiled for collaborators who use ALSWH data

ALSWH sampling scheme for the 1921-26, 1946-51 and 1973-78 cohorts

 

Selection of the sample

1973-78, 1946-51 and 1921-26 cohorts

The study sample was selected by Medicare Australia (previously known as the Health Insurance Commission) from three zones - urban, rural and remote defined according to RRMA(ref) where urban includes Capital City and Other Metropolitan Centres; rural, Large Rural Centres, Small Rural Centres and Other Rural Areas; and remote, Remote Centres and Other Remote Areas. The age groups sampled from the Medicare database in April 1996 were 18-22 years, 45-49 years and 70-74 years. By the time the invitations to participate were mailed later in 1996, some women at the upper limit of the age groups had had their birthday and were a year older. Hence some women recruited were 23, 50 and 75 years old and so the cohort age ranges in the study are: 18-23; 45-50 and 70-75 years (although there are relatively fewer women in the oldest year of each cohort). The cohorts are now referred to by their years of birth but some ALSWH material may refer to them as 'Young', 'Mid-aged' and 'Older' and data sets use 'y', 'm', and 'o' (further information below). Sampling from the population was random within each age group, except that women from rural and remote areas were selected in twice the proportions of the Australian population living in these areas. Women from capital cities and other metropolitan areas made up the balance of the samples. There were also a small number of women invited to participate whose age was outside the cohort birth years (by a year or two), possibly due to errors in date of birth in the Medicare database. However the survey data for these women have been retained. We recommend that when using the data, these women are either excluded or their age set to the nearest valid age.

1989-95 cohort

Please note that some variables in Surveys 1 and 2 of the 1989-95 cohort were renamed for consistency in April 2016.
Renaming of Variables in the Surveys 1 and 2 for the 1989-95 cohort

Recruitment for the 1989-95 cohort was different from the other cohorts.  A variety of recruitment strategies were used (see the Report I, section 3.)   A brief summary is given here.

For inclusion in the 1989-95 cohort, respondents needed to:

  • meet the eligibility criteria of being female, aged 18 to 23 and having a Medicare number;
  • answer at least some survey questions;
  • meet the requirements for data linkage.

A total of 17,567 women met the above inclusion criteria. To establish a pilot study group for the cohort, the first 498 young women that met the above criteria were removed from the main cohort. As a result the pilot study group included all women recruited in October 2012 who were verified by the Department of Human Services. Of the remaining sample, 17,069 participants were verified by the Department of Human Services.

 

Calculation of the sample weights

1973-78, 1946-51 and 1921-26 cohorts

The women were selected based on their postcodes recorded by Medicare. The variable in the datasets called 'inarea' reflects the area from which the women were sampled (urban, rural, remote). However by the time the survey was mailed, some women, particularly in the younger age group, had moved. The variable 'y1area' reflects their actual area of residence when completing the survey. The number of respondents who lived in urban, rural and remote areas at the time of completing the first survey in 1996 (wave 1 area) was used to create the sample weights for each age group for each area (urban, rural, remote), by comparing these numbers of respondents to 1996 Census figures. The sample weights appear in the datasets and are labelled y1wtarea, m1wtarea and o1wtarea.

1989-95 cohort

For the 1989-95 cohort, the Census was used as the best available measure of Australia’s population of women aged 18 to 23, despite including a considerable number of overseas students and temporary residents who are ineligible for Medicare (and were thus not included in the ALSWH cohort). In order to compensate for the over-representation of women from educated backgrounds and the under-representation of 18 year olds in the sample, weightings were calculated using the following formula:

Weights for women in the sample of age x (at baseline) with education level y:

Where N is the total number of women in the sample and N(x,y) is the number of women aged x years with education level y in the sample. Similarly P is the total number of women aged 18 to 23 in the Australian population, and P(x,y) is the number of women in the Australian population aged x years with education level y.

Those with missing data recorded for their educational qualifications (as was the case for 7.8% of the women in the population) were omitted from the calculation of weights, which in effect assumes that the data are missing at random.

 

Representativeness and attrition


The following articles are the best references for current retention rates and representativeness:

Lee C, Dobson AJ, Brown WJ, Bryson L, Byles J, Warner-Smith P, Young AF. (2005) Cohort Profile: The Australian Longitudinal Study on Women's Health. International Journal of Epidemiology; 34: 987-991.

Young AF, Powers JR, Bell SL. Attrition in longitudinal studies: who do you lose? Australian and New Zealand Journal of Public Health. 2006 Aug;

Brilleman SL, Pachana NA, Dobson AJ. The impact of attrition on the representativeness of cohort studies of older people. BMC Medical Research Methodology. 2010 Aug;10.

Powers J, Loxton D. The Impact of Attrition in an 11-Year Prospective Longitudinal Study of Younger Women. Annals of Epidemiology 2010;20(4):318-21.)

For representativeness for the 1989-95 cohort see:

Health and wellbeing of women aged 18 to 23 in 2013 and 1996: Findings from the Australian Longitudinal Study on Women’s Health. Mishra G, Loxton D, Anderson A, Hockey R, Powers J, Brown W, Dobson A, Duffy L, Graves A, Harris M, Harris S, Lucke J, McLaughlin D, Mooney R, Pachana N, Pease S, Tavener M, Thomson C, Tooth L, Townsend N, Tuckerman R and Byles J. Report prepared for the Australian Government Department of Health, June 2014.  (Section 4).

 

Longitudinal analysis

When doing longitudinal analyses, remember to weight for area of residence at Survey 1 (y1wtarea, m1wtarea, o1wtarea) in all crosstabs, frequencies and analyses to adjust for the initial deliberate oversampling in rural and remote areas. This weighting may not be required in models that include area of residence.

 

Missing data

Some participants completed a short survey instead of the full survey, accounting for some missing data. The type of survey completed is identified with variables such as y2survey for Survey 2 of the 1973-78 cohort. Survey 2 of the 1946-51 cohort Q70 on income is missing the first category ($1-$119). There are large amounts of missing data in some income questions. Surveys 2, 3 and 4 of the 1946-51 cohort are missing the question about being admitted to hospital. Survey 2 of the 1973-78 cohort is missing the question about ability to manage on income. Survey 2 of the 1946-51 cohort Q67 is unreliable as the instruction was incorrectly stated as "mark one only" rather than "mark all that apply". Many participants realised that this was an error and answered the question as it should have been. Others may not have done so.

The first survey of the 1989-95 cohort has 167 records whose data are almost all missing.  These records are identified by the allmissing variable. This variable has the value 1 for those records that are almost all missing, zero otherwise.  These records represent eligible respondents who did complete the first survey but we unfortunately lost their data.  They are kept in the data set so that the first wave’s data set contains the whole sample.

Notes about data files

The quantitative survey data are available as SAS, STATA and SPSS data files, or as tab delimited text files. The file includes almost all survey items as well as all derived and calculated variables.

As well as the survey datasets, there are some supplementary datasets that have been created. Some of these require a written request for the data (e.g. FFQ data). Information about dates of deaths and withdrawal of participants is available in the participant status file.

The qualitative data recorded on the back page are also available for analyses. For further information refer to the Qualitative processing protocols at www.alswh.org.au/how-to-access-the-data/alswh-data. 
For more information about using ALSWH data and applying to the Publications, Sub-studies and Analyses Committee for access to the data please refer to the website:alswh.org.au/how-to-access-the-data/alswh-data

 

Extra resources to support data analysis

Check the data map, the data dictionary and Data Dictionary Supplement for further information about survey items and derived variables. They are available by following this link.

The Data Dictionary is a Microsoft Access database that gives a detailed description of the questions used in the survey, their source and how they are used, as well as information on the derived and calculated variables. The Data Dictionary is constantly updated and is available here. (The table is over 1,000 pages long so don't try to print it).

The Data Dictionary Supplement is a series of documents that accompanies the Data Dictionary. The Data Dictionary Supplement contains information about scales and other measures used in the ALSWH surveys. Before using any summary or scale score included in an ALSWH dataset, the appropriate section of the Data Dictionary Supplement should be reviewed. The Data Dictionary and Data Dictionary Supplement are available here.

Check the survey data books if unsure about response frequencies. Electronic copies of the surveys and data books are available here.

 

More information about quantitative survey data files

In general it is the responsibility of the analyst to become familiar with and carefully examine all data before proceeding with data analysis.

There are different naming conventions for survey items and derived items. IDalias is a unique de-identified participant number, present in all data files. This participant number can be used to merge data files across surveys. The survey questions and method used in the calculation of the derived variables are listed in the Data Dictionary. A few survey items at Survey 1 (birth date, country of birth, language spoken at home) were removed or aggregated into groups as these were considered potentially able to make participants 'identifiable'.

It is not recommended to arbitrarily replace missing values with the null value or any other value. Questions involving "mark all that apply" responses have been coded to 0 (no response) or 1 (yes response). In general, a "none of the above" response option was offered at the end of each set of "mark all that apply" questions. If responses to all sections of a specific question were missing, including the null option ("none of the above"), all responses were set to missing.

 

Naming conventions for datasets

The datasets are named whanaaaB.txt, where:

n = survey number

aaa = agegroup – yng (1973-78 cohort), mid (1946-51 cohort) , old (1921-26 cohort) or nyc (1989-95 cohort)

B = level B data (identifying information removed).  E.g., wha1yngB.txt is text data for Survey 1 of the 1973-78 cohort.

Naming conventions for variables

The variables are named anvarname where:

a = agegroup – y(1973-78 cohort), m (1946-51 cohort) , or o (1921-26 cohort).

N = survey number

The 1989-95 cohort (also referred to as the New Young Cohort, or NYC) has been allocated  the one-letter abbreviation ‘z’ because it follows on from the first young cohort, which used ‘y’.  However, the variable names in the 1989-95 cohort data do not use the prefixes used in the other cohorts.

 

Survey variables

varname is the question number in the survey, e.g., q1, q34a.

The variables for the 1989-95 cohort do not have questionnaire numbering or prefixes.  The Data Dictionary has a list of all the variable names.

 

Derived and calculated variables

Varname is a descriptive name, e.g., pcode, cesd10.  Prefixes are as follows:

m1 = survey 1 of the (mid) 1946-51 cohort

o1 = survey 1 of the (older) 1921-26 cohort

y1 = survey 1 of the (younger) 1973-78 cohort

m2 = survey 2 of the (mid) 1946-51 cohort

o2 = survey 2 of the (older) 1921-26 cohort

y2 = survey 2 of the (younger) 1973-78 cohort, etc.

 

Associated Documentation files

 

Notes about specific variables

Child data set – A set of questions relating to child birth have been included for the 1973-78 cohort from the fourth survey onwards. The data from these child-specific questions are included in a separate data set for each survey.

Items that form part of a scale – Be careful that you do not inappropriately analyse single items from a scale. For example, the 36 items in the SF-36 should not be considered as separate items, other than the first self-rated health item. The Data Dictionary Supplement has details about which scales have been included in the surveys.

Counting symptoms - when looking at symptoms, the general rule is to count the number of women who had the symptom "sometimes" or "often".

Measure of depressive symptoms - the 10-item CES-D scale has an extra item at the end ("I felt terrific") which is not included in the calculation of the CES-D score. The CES-D score is available in the datasets.

Menopause - The menopause status variable is recalculated as each new dataset becomes available for the 1946-51 cohort. Make sure you get the most recent menopause status dataset.

Medications data sets – The fourth survey of the 1921-26 cohort, the fifth and sixth of the 1946-51 cohort and the fifth and sixth of the 1973-78 cohort have data on self-reported medications the respondents are taking. These data are available on separate data sets. Where possible, the medications are given by name and ATC code.

Measures of physical activity - the physical activity questions were changed after Survey 1. The new physical activity measures from Survey 2 are not comparable to Survey 1 in longitudinal analysis. Refer to the Data Dictionary Supplement for more information.

Summary variables - there are a few "standard" ways to collapse some of the main categorical variables we collect. For example, education (highest qualification) can be dichotomised as "school only", "post school" or in three categories: "no formal qualifications", "school qualifications", "trade/tertiary qualifications" and so on. There have been several variables created to summarise sets of items in the surveys (eg. the illicit drug use items) and it is important that data analysts become familiar with these new variables (See Data Dictionary Supplement)

Area of residence - the recommended measure is ARIA+. This is an index of accessibility/remoteness based on the distance to the nearest service centre. The scores range from 0 to 15 and the ABS has defined 5 categories for remoteness: major cities of Australia, inner regional Australia, outer regional Australia, remote, and very remote. Only a few of the study's women live in very remote areas, so the fourth and fifth categories are often grouped together. Aria+ is recommended over the previously used RRMA area classification. See http://www.adelaide.edu.au/apmrc/research/projects/category/aria.html for more information.

ATSI status - asked at Survey 1 in all age groups. This variable can be used in statistical models but results should not be reported separately by ATSI status in any reports. See http://www.alswh.org.au/for-researchers/data/indigenous-policy for more information.

 

Participant Status and Cause of Death

For a detailed description of Participant Status and Cause of Death files please see section 8 of the Data Dictionary Supplement page.

 

Short Surveys in the ALSWH

Shorter questionnaires have been used for some respondents in ALSWH when the women had not responded and was contacted late and offered a short survey to complete.  The short surveys were only offered in the second surveys of the 1921-26, 1946-51, and 1973-78 cohorts, and the third survey of the 1946-51.

The short surveys only contained those questions that were considered particularly important.   These questions are listed in the Short Surveys document.  The researcher can identify which respondent did the short survey because their ‘survey’ variable will have the value 2 rather than 1.    These records will have many variables that are entirely missing; the variables that were not included in the short survey.