Thread by @rnishimura

Tweeting now session "Come In and Stay a While: Recruiting and Measuring Attrition" at #AAPOR

Kyle (SSRS) starts the session talking about chronic-nonrespondents in the SSRS probability panel

Defining chronic nonrespondents -- people who have not responded the survey after at least six invitation.

Principals of re-engaging chronic nonrespondents
High-incentive: up to $20
Promise of short "5 questions survey"

Last chance to continue to be part of the panel

How to track chronic nonrespondents?
Do they actually continuing answering the survey or do they drop (hit-and-run)

55.3% completed 1 or more surveys after re-engagements.

Next:

Kyle continues the session to compare non-internet and web-reluctant panelists. Some concerns regarding under-coverage here.

Multi-mode vs Internet enabling approach
Pros and Cons with each approach:
Internet enabling is expensive at front cost, but cheaper later
Multi-mode allows respondents yo answer on their mode of preference, but measure error differences.

Asked web-relunctant panelists why they prefer taking the survey over the phone instead of online.

I would add: because depending on the topic, they should be concerned about coverage error

The paths panelists of the SSRS panel can answer the surveys:

66% of the phone panelists says they prefer to continue taking the survey over the phone
54% for web-relunctant
91% for non-internet

Non-internet panelists are older and less educated than web-relunctant panelists

Summary/Take-aways
Non-internet users are different than web-relunctant users
Choice of mode allows to panelists answer the survey on the mode of their choice and capture non-internet users

Next is Fei He (Ipsos) talking about innovations in recruiting non-internte households in KnowldgePanel

6% adults in the US don't have internet at home. They then to ve older, less educated, low-income and Hispanic.

KnowledgePanel recruitment by a mail sample recruitment with a pre-paid incentive.
For the non-internet panelists, they are provided internet.
Older methodology: Dial-up service with notebooks
Nee methodology: Tablet with a mobile plan

Panelists become active quicker with new methodology

Table has lower attrition than dial-up service with laptop

Next is Kevin Tolliver (US Census Bureau) talk about identifying key factors that lead to attrition in US Census Bureau demographic surveys

Analyzed the following surveys:
American Household Survey (AHS)
Current Population Survey (CPS)
National Survey of College Graduates (NSCG)
Survey of Income Program Participation (SIPP)

Four very different surveys!

Tested various characteristics relation to attrition with Wald Chi-square test and corresponding odds ratio

Based on these results, built attrition prediction models.
Objective is to predict attrition in one survey based on data of other survey

Then, these prediction models can be used to identify specific groups that have higher attrition propensity

Attrition prediction models were all not great 😭

And that's a wrap for this session
/TheEnd

@threadreaderapp compile unroll thread