Thursday, February 24, 2011

AAPOR Report on On-Line Panels


 

Type of Online Panels

-          Vast majority not constructed using probability based recruitment

-          Offers to join

-          River sampling

 

Total Survey Error

-          Coverage error – major factor when goal is to represent general population

-          Approx 1/3rd Us adult population does not use internet on regular basis.

-          Under coverage

-          High level of non response at various stages of building a non-probability panel and delivering respondents to individual studies

-          Major differences between surveys using non probability panels and traditional methods (usually phone) à difficult to determine whether mode of administration or sample bias is greater cause of differences.

-          Studies suggest probability sampling still more accurate

 

Adjustments to Reduce Bias

 

-          Standard demographic weighting

-          Simple purposive sampling that uses known information about panel members to generate demographically balanced samples

-          Standard quota sampling

-          Propensity models in post stratification adjustments à augment standard demographic weighting with attitudinal or behavioural measures thought to be predictors of bias

 

 

Concerns about Panel Quality

 

-          Industry response

o   Guidelines & standards

o   Validating panelists à remove duplicates etc

o   Research to understand drivers of panelist behaviours / design techniques to reduce impact of those behaviours on survey results.

 

 

Background & Purpose of Report

 

-          Types of panels

o   Probability based methods

o   Non probability approach / volunteer online panels

 

 

Overview

 

-          First challenge for all survey modes à development of sample frame

-          Online panels have become popular solution to sample frame problem.

-          General population panel à serves as a frame from which samples are drawn to meet specific needs of particular studies.

-          Census balanced samples

-          Specialty panel

-          Proprietary panel

-          Targeted samples

 

 

Non Probability Volunteer Online Panels

 

 

-          Lower cost

-          Faster response

-          Ability to build targeted samples of people who would be low incidence in a general population sample

-          Five major areas of activity

o   Recruitment of members            

§  Motivations

·         Contingent incentive

·         Self expression

·         Fun

·         Social comparison

·         convenience

o   Joining procedures and profiling

§  Double optin process

§  Validation procedures

o   Specific study sampling

§  Simple random samples from panels are rare à tendency to be highly skewed to particular demographics

§  Purposive sampling

o   Incentive programs

o   Panel maintenance

 

 

Probability Based Recruitment

 

-          Dutch Telepanel – 1986

 

 

River Sampling

 

-          Intercept interviewing / real time sampling

-          May be on rise as researchers seek larger / more diverse sample pools and less frequently surveyed respondents than provided by online panels

 

Errors of Non Observation in Online Panel Surveys

 

-          Coverage

-          Non response

-          More severe with online panels than other types of surveys

-          All sampling frames have features which can affect quality

o   Under coverage

o   Multiple mappings of frame to population

§  Eg, multiple people per home phone line

o   Duplication

§  eg, one person, multiple phone numbers

 

Online Panel Surveys, Frames, and Coverage Issues

 

-          volunteer panels do not attempt to build complete sampling frame

-          notion of sample frame is skipped

-          instead focuses on recruitment and sampling steps.

-          Common evaluative criterion of volunteer panel is not full coverage of household population, but sufficient diversity on attributes related to type of surveys supported by panel.

-          Online panels can repeatedly sample

 

 

Unit Non Response and Non Response Error

 

 

-          Failure to measure a unit in a sample

-          Person selected for a sample does not respond to survey

-          Vs item non response à respondent skips a question

-          4 stages in volunteer panels where non response can be a issue

o   Recruitment

§  No way of knowing anything precise about size / nature of non response at this stage

§  If online panel members belonging to under represented groups are similar to group members who are not in panel, the risk of bias is diminished under an appropriate adjustment procedure

·         Within group homogeneity may be a poor assumption.

o   Joining and profiling

§  Just over 6% of those who click thru a banner advert to panel registration page eventually become member.

o   Specific study sampling

§  Reasons why member may not participate

·         Lack of interest / survey length / heavy volume of survey invitations

·         Failure to qualify / not completing within require time period

·         Technical problems

§  Address differential non-response

·         Disproportionately higher rate of group selected.

·         Pre emptive differential non response adjustment

·         No guarantee that non response error eliminated / reduced

o   Panel maintenance

§  Forced or normal attrition

§  Forced turnover is not a form of non response

-          Response metrics

o   Recruitment is constant on-going endeavour

-          Coverage errors versus non response bias

o   Given absence of sampling frame in online panels, conceptual difference between coverage errors and non response errors gets blurred.

-          Measurement error in online panels

o   Understand how and why people think, feel and act.

o   Measurement error is defined as difference between observed response and underlying true response

§  How concepts are measured

·         The questions and the answers

·         Questionnaire design effects

§  Mode of interview

§  Respondents

§  Interviewers

o   Mode effects

§  Two hypotheses about possible impact of shifting from one mode to another

·         Social desirability

·         Satisficing

§  Primacy

·         Tendency for respondents to select answers offered at beginning of list

§  Recency

·         Select answers from last offered

§  Concurrent – predictive validity

o   Respondent effects

§  Cognitive capabilities of panel members

§  Motivation to participate

§  Panel conditioning

·         Taking too many surveys

§  Topic interest and experience

-          Sample adjustments to reduce error and bias

o   Purposive sampling

§  Non random selection technique

§  Goal – sample representative of defined target population

§  Anders Kiaer – 19th century

§  Quota sampling

§  Census balance sample

o   Model based methods

§  Small area estimation

§  Epidemiological studies

o   Post survey adjustment

§  Weighting techniques

-          Panel data Cleaning

 

-           

 

No comments: