- CAWI means a web questionnaire completed by the respondent, usually on an online panel of people recruited and verified in advance.
- It is the right method for concept tests, campaign evaluations, studies among customers with e-mail addresses and connected urban segments.
- An online panel describes active internet users well and everyone else badly; for the general population, representativeness must be demonstrated, not assumed.
- Answer quality depends on attention checks, duration control, duplicate detection and the removal of professional respondents.
- An online survey takes 3-10 days and costs little; the saving is real only if the target population really is online.
1. What CAWI is and what an online panel is
CAWI (Computer-Assisted Web Interviewing) is the online survey: the respondent receives a link, opens the questionnaire in a browser or on a phone and completes it alone, without an interviewer. The questionnaire is programmed with the same filters and validations as CAPI and CATI, plus possibilities the other methods lack: images, video clips, long searchable lists, drag-and-drop ranking exercises, choices between product variants displayed side by side.
The essential question is not how the questionnaire is completed but who receives it. There are three sources of respondents. The online panel is a base of people recruited in advance, with a verified demographic profile, who have agreed to take part in studies periodically for a small reward; the agency selects from the panel the respondents who match the study's criteria. The client's own list, for example the e-mail addresses of customers or of an organisation's members, allows an online survey of a known population. The open link, distributed on social networks or a website, is the weakest source: nobody knows who answered, who answered several times and who was sent to answer.
2. Which studies the online survey is the right method for
- Concept, packaging and advertising tests. The material must be seen, variants can be allocated at random, and the target population, urban consumers of the category, is usually online. The guide on monitoring and testing describes these studies.
- Campaign evaluations and awareness tracking among urban segments, with frequent waves and a short questionnaire.
- Customer studies with e-mail addresses: satisfaction, usage, renewal intention. The base is known, selection can be random.
- Employee studies, where the list is complete and online anonymity increases candour.
- Studies of specific segments that are hard to find in the general population: users of an app, buyers of a niche category, professionals in a field. The panel allows them to be filtered quickly.
- Quick studies, where the result is needed within days and a good approximation is worth more than a perfect measurement a month later.
3. The representativeness problem: who is not in the panel
An online panel consists of people who have internet access, use it frequently, agreed to join a panel and complete questionnaires for a reward. Each of these conditions removes part of the population, and the part removed is not random: older people, rural residents, those with lower education and income, and those who simply have no time or interest in questionnaires are under-represented. In the Republic of Moldova, where differences in internet access between urban and rural areas and between age groups remain significant, this distortion is not a detail.
Weighting corrects the structure of the sample by gender, age, region and education so that the proportions match the population. What weighting cannot do is change who the people in each cell are. A 65-year-old rural woman who completes online questionnaires is not representative of 65-year-old rural women; she is representative of those who are connected and active online, a minority with opinions and behaviours of their own. The more the subject of the study relates to access to technology, information or services, the greater the distortion.
For studies that require coverage of the whole population, the reference method remains the face-to-face interview. A frequent combination is a mixed design: CAPI for the main sample, with an additional CAWI module on the young urban segment, where online is more efficient and coverage is not a problem.
4. Answer quality: the unsupervised respondent
Without an interviewer, nobody sees whether the respondent reads the questions, answers seriously or is even the person answering. A good online survey compensates through programmed mechanisms, and a CAWI proposal must describe them.
| Mechanism | What it removes | How it works |
|---|---|---|
| Identity verification at recruitment | Fake profiles, people outside the population | Confirmation by e-mail and telephone, demographic data checked over time |
| Duplicate detection | The same person several times | Device fingerprint, IP address, cookie, panel check |
| Attention checks | Respondents who do not read | Instructions hidden in the text ("choose the third option"), questions with a known answer |
| Duration control | Mechanical completions | Removal of those below 40-50% of the median duration |
| Answer patterns | The same option on every line, zigzag answers | Automatic analysis of scale batteries |
| Open-answer review | Meaningless, copied or generated text | Manual reading of a sample, detection algorithms |
| Participation limits | Professional respondents | Maximum number of studies per month, exclusion of those with abnormal qualification rates |
In a well-run study, between 5% and 15% of completed questionnaires are removed by these checks. A proposal that mentions no removals does not mean the panel is perfect; it means nobody checked.
5. The online questionnaire: short, visual, on the phone
Most online questionnaires are completed on a phone, and this imposes strict rules. The optimal length is 8-12 minutes, and beyond 15 minutes the drop-out rate rises sharply. Large scale grids, readable on a computer screen, become unreadable on a phone and are broken down into individual questions. Images must be clear at small sizes. Each screen holds a single question or a short battery, and the progress bar is visible.
The specific advantage of the online survey is random allocation: respondents can be split automatically into groups that see different variants of a concept, a price or an advertisement, without knowing the other variants exist. This is the foundation of comparative tests and price experiments, impossible to run as cleanly by telephone and expensive face to face.
6. Sample size and reading the results
An online study of the urban population usually uses samples of 500-1,200 respondents, sized by the reporting subgroups rather than the total. For tests with random allocation, each variant needs 150-300 respondents so that differences between variants can be tested statistically. The margin of error is calculated as for any sample, but reported with a caveat: it describes the precision of the estimate for the panel population, not for the general population. The methodology FAQ explains the calculation.
When reading the report, check three things: the structure of the achieved sample before and after weighting, the size of the extreme weights, and the share of respondents removed by quality control. If weights exceed values of 3-4, certain groups were so scarce in the panel that a few respondents "speak" for many, and the results for those groups are fragile.
7. Online, telephone or face to face: a decision by population
| Target population | Suitable method | Reason |
|---|---|---|
| General population 18+, including rural and 60+ | CAPI | The only complete coverage |
| Urban population 18-45 | CAWI, possibly with CATI | Good online coverage, low cost |
| A company's customers, with e-mail | CAWI | Known base, random selection, anonymity |
| A company's customers, with telephone | CATI | Higher response rate than by e-mail |
| Buyers of a niche category | CAWI on a panel | Fast, inexpensive filtering |
| Beneficiaries of a social programme | CAPI | Identified in the field, limited internet access |
| An organisation's employees | CAWI | Complete list, anonymity |
8. What to ask the agency for an online survey
- The source of respondents: own panel, partner panel, the client's list or an open link.
- The size of the active panel and its demographic structure against the target population.
- How panel members are recruited and verified.
- The quality control mechanisms and the estimated share of questionnaires removed.
- The questionnaire length tested on a phone, not only on a computer.
- The weighting variables, the reference source and the range of the weights.
- The exact wording in which the results may be presented: for which population they are valid.
The online survey is the data collection method that has changed market research most in the last decade, and rightly so: for connected populations it is fast, inexpensive and allows experimental designs the other methods do not. The problem was never the method, but its use for populations it does not cover. The question "can't we just do it all online?" has a simple answer: you can, if everyone who matters for your decision is online. If not, the low cost of the online survey is the high cost of a decision taken on part of the population. The alternative methods are described in the guides on the telephone survey and the face-to-face interview.
See also: CAWI Online Panel Quantitative research Monitoring and testing Face-to-face interviews (CAPI) Telephone surveys (CATI) Methodology FAQ
Do you have a decision to take?
Tell us the problem, not the method. We propose the right research design and explain what it can and cannot deliver.