AI Glossary / Statistics and probability / Intermediate

Computer Assisted Survey Information Collection

Definition

Computer Assisted Survey Information Collection, usually abbreviated CASIC, is an umbrella term for using computers to support survey data collection and capture. It covers more than one interviewing mode and can include related preparation and operational activities. The technology helps administer questions, record answers, and manage information, while the survey still needs a sound design, clear questions, and appropriate treatment of respondents.

What it means

A survey is more than a list of questions. It includes selecting whom to contact, deciding which questions apply, recording responses accurately, and preparing usable data. CASIC brings computer support into these activities.
Examples include computer-assisted telephone interviewing, where an interviewer enters responses during a call; computer-assisted personal interviewing, where an interviewer uses a device in a face-to-face setting; and computer-assisted self-interviewing, where respondents complete an electronic instrument themselves. A web questionnaire is one possible implementation. CASIC does not require artificial intelligence, an autonomous interviewer, or a continuously connected device.

How it works

Start with the survey's purpose, target population, sampling plan, and information requirements. Translate the questionnaire into a tested electronic instrument. Define response types, routing rules, consistency checks, and how refusals, uncertainty, or unavailable answers are recorded.
During collection, the system presents relevant questions and stores responses. It may insert earlier answers into later wording, calculate totals, or flag an apparent inconsistency for clarification. Such features must be designed carefully: a warning should not pressure someone into replacing a valid unusual answer with a more convenient one.
Supporting functions can include assigning cases to interviewers, tracking contact attempts, saving interrupted interviews, and transferring records. The exact functions depend on the survey and implementation. Test navigation after an earlier answer changes, because downstream questions may then become inapplicable.
Before launch, assess question comprehension and usability with the intended users, including interviewers where relevant. Monitor missingness, routing errors, device problems, and differences between collection modes. Maintain clear versions of the instrument and data definitions. Computer support can improve consistency, but it can also reproduce a programming mistake across every affected interview.

Three practical examples

• A field interviewer records answers on a tablet, with the instrument displaying only questions relevant to the respondent's earlier answers.
• A telephone survey system guides an interviewer through a standardized questionnaire and records contact outcomes separately from answers.
• A customer research survey lets respondents complete an electronic questionnaire themselves, preserving distinctions between skipped, unanswered, and explicitly declined items.

Think of it this way

Think of CASIC as a guided survey workbench. It organizes the steps, keeps records, and can draw attention to inconsistencies. It does not decide whether the people invited represent the population, whether a question is meaningful, or whether an answer reflects the respondent's experience. Those remain design and interpretation responsibilities.

Worked example

Consider an original routing example for a business survey. Every completed interview contains 10 core questions. An additional eight questions apply only to businesses that used a particular service. Suppose 300 interviews are completed and 120 businesses report using it.
Correct routing displays 300 times 10 = 3,000 core questions and 120 times 8 = 960 follow-up questions, for 3,960 question presentations. Showing all 18 questions to everyone would produce 300 times 18 = 5,400 presentations. Routing avoids 1,440 inapplicable presentations, or 1,440/5,400 = 26.67% of that unrouted total.
Now suppose a programming error sends 15 nonusers into the follow-up section. It creates 15 times 8 = 120 inappropriate presentations. The error affects 15/300 = 5% of completed interviews, but that percentage is not a survey response rate and does not measure every type of data error.
The arithmetic illustrates instrument behavior, not guaranteed time savings. Question duration varies, and legitimate follow-up clarification can add time. The team must also verify that eligible users were not incorrectly routed away from needed questions.

Illustrative case study

In a fictional organization, Willow Insights moves a small-business survey from paper to a computer-assisted instrument. The team initially assumes that digitizing the questions will be enough. Pilot interviews reveal that changing an early answer leaves some later fields populated even though they no longer apply.
The team corrects the routing and defines how obsolete answers are handled. It tests unusual but valid responses, interrupted interviews, and respondents who decline an item. Interviewers receive practice with the revised instrument before fieldwork begins.
Managers then review completion patterns and data quality alongside operational effort. This fictional example illustrates testing and monitoring; it does not report real savings or imply that electronic collection automatically outperforms paper.

Common misconceptions and mistakes

CASIC is not synonymous with an online poll or with AI interviewing. Automated validation cannot establish that an answer is true. A mandatory field may reduce blanks while increasing inaccurate answers or abandonment. Faster capture does not remove coverage error or nonresponse bias. Different modes can influence how people understand or answer questions. Copying a paper layout onto a screen without usability testing can preserve old problems and introduce new ones.

Business use

Businesses use CASIC for customer research, employee feedback, supplier studies, and structured field assessments. It can connect question delivery with consistent recording and operational monitoring. Evaluate success using relevant measures such as routing accuracy, respondent burden, usable data, and total collection effort. If AI assists with transcription or coding, assess those functions separately and preserve the distinction between the respondent's answer and a model's interpretation.

People also ask

Must an interviewer be present?

No; both interviewer-administered and self-administered approaches can fall within CASIC.

Does the computer choose the sample?

It may support sampling or case management, but the collection technology does not itself define a valid sampling design.

Can electronic checks eliminate errors?

No. They can detect specified problems, while misunderstanding, omission, and programming errors still require attention.

Check your understanding

In the worked example, how many follow-up presentations are appropriate?

Answer: There are 960.

Does avoiding 26.67% of presentations prove the survey is 26.67% faster?

Answer: No, because questions and interactions take different amounts of time.

Why distinguish a skipped item from a refusal?

Answer: They have different meanings: one may be inapplicable by design, while the other records a respondent's decision not to answer.

Related terms

• Nonresponse
• Computer-assisted telephone interviewing
• computer-assisted personal interviewing
• computer-assisted self-interviewing
• questionnaire routing
• survey instrument
• data capture
• mode effects
• sampling frame
• paradata

Learn next

Learn questionnaire design and survey error before adding complex automation. Then study routing tests, interviewer usability, and the interpretation of missing responses. For projects involving AI, examine how transcription and coding accuracy can be assessed without treating generated interpretations as original respondent statements.

More in Statistics and probability

• Component analysis (statistics) - Component analysis, in the treatment-evaluation sense used here, systematically investigates the elements of an intervention package to understand their...
• Compound symmetry - Compound symmetry is a covariance structure in which all measurements have the same variance and every distinct pair has the same covariance.
• Computational statistics - Computational statistics develops and applies algorithms for statistical analysis, especially when estimates, uncertainty measures, or model calculations...
• Concordant pairs - Concordant pairs are pairs of observations whose ordering agrees on two variables.
• Conditional distribution - A conditional distribution describes how a variable is distributed when specified information about another variable or event is given.
• Conditional means - Conditional means are the expected values of a variable for specified values or conditions of another variable or set of variables.

Sources

OECD Glossary of Statistical Terms, Computer Assisted Survey Information Collection: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf Cathryn Dippo, Designing an instrument for computer-assisted data collection in the Current Population Survey, BLS-hosted research: https://www.bls.gov/osmr/research-papers/1993/pdf/cp930010.pdf William Mockovak, Survey Data Collection Using Complex Automated Questionnaires, BLS-hosted research: https://www.bls.gov/osmr/research-papers/2003/pdf/st030330.pdf

Previous: Compute Next: Computer audition