Methods to account for self-selection bias

Location: ABS Office, Nationally

Duration: 4 months

Proposed start date: April 2019

Project Background

Traditionally, the ABS has conducted probability samples where each respondent to a survey has a known probability of selection.  This probability of selection is used in weighting the survey data to produce reliable estimates for the population.

In seeking to utilise data sources beyond probability samples, there is a need to deal appropriately with “self-selection bias” – the bias that results when respondents do not have a known probability of selection, due to the ability to self-select into the sample for a survey.  An example is online panel data, where participants may voluntarily choose to participate in an online survey, which may be for a small reward.

Another example is for ABS household surveys where a single person is selected from within a household to answer personal questions.  Previously a household interviewer would randomly select a person from within the household.  Using online surveys it may be simpler to allow a household to self-select a person from within the household.  Even if a person from within the household is selected randomly as part of an electronic form, there is still a possibility that a household will ignore this and instead self-select a household member to respond.

A further example is where households selected for a survey can self-select their mode of response, e.g. they can choose to respond via online reporting or through a personal interview.  If we wanted to include some questions only on the online form (which may be advantageous for some sensitive questions), then the sample for these online questions would again be self-selected.

The research question is what methods can be used to account for self-selection bias and use samples that include self-selected respondents to produce reliable estimates for the full population.

Research to be Conducted

Conduct a literature review to determine a range of methods that may be considered to account for self-selection bias in contexts relevant to the work of the ABS.  In consultation with the ABS supervisor, develop suitable methods and trial them on appropriate datasets.  As access to actual survey data with self-selection effects may be limited, the trialling of methods may be performed on simulated datasets, looking at a range of possible underlying self-selection models.

Skills Required

We are looking for someone with the following skillset:

  • Good understanding of survey sampling and survey estimation
  • Statistical computing skills appropriate for generating a simulated population, and for selecting samples from within a population, that are generated from underlying statistical models
  • Knowledge of ABS surveys and ABS survey data (such as ABS Confidentialised Unit Record Files) is an advantage but not essential

Expected Outcomes

This project will assist the ABS to choose suitable methods for dealing with self-selection effects for future scenarios.

Additional Details

The intern will receive $3,000 per month of the internship, usually in the form of stipend payments.

It is expected that the intern will primarily undertake this research project during regular business hours, spending at least 80% of their time on-site with the industry partner.  The intern will be expected to maintain contact with their academic mentor throughout the internship either through face-to-face or phone meetings as appropriate.

The intern and their academic mentor will have the opportunity to negotiate the project’s scope, milestones and timeline during the project planning stage.

Applications Close

03 April 2019


APR – 0890