The decision whether to conduct a quantitative study online, by telephone, or through face-to-face interviews determines who will actually be included in the sample and how respondents will answer sensitive questions. The differences between CAPI, CATI, and CAWI are not merely technical but methodological – each of these techniques changes the structure of respondent reach, questionnaire length, the quality of stated responses, and the level of interviewer effect. Below is a breakdown of how the respondent contact channel affects the reliability of results.<\/p>\n
The three main quantitative data collection methods – CAPI (Computer-Assisted Personal Interviewing), CATI (Computer-Assisted Telephone Interviewing), and CAWI (Computer-Assisted Web Interviewing) – differ not only in the tools used, but above all in how they reach and engage respondents. The differences between CAPI, CATI, and CAWI become apparent across four dimensions: the structure of the reachable sample, the length and complexity of the questionnaire, respondents’ willingness to provide honest answers, and the presence or absence of an interviewer as a mediator.<\/p>\n
CAPI is a face-to-face interview conducted by an interviewer using a tablet or laptop, most often in the respondent’s home or in a public place. CATI is a telephone survey conducted from a studio or remotely, with the interviewer entering responses into the system. CAWI is an online survey completed independently by the respondent – through a link in an email, an invitation from a research panel, or an invitation in an app. Each of these channels favors a different type of participant and makes it more difficult to reach other groups.<\/p>\n
In practice, the difference is not about asking “which method is better,” but rather “which method distorts results in a way that is acceptable for a given research problem.” In studies of voting intentions, telephone and online surveys may produce different distributions than face-to-face interviews. In research on alcohol consumption, CAWI generally produces higher reported levels than CAPI because respondents do not have to look the interviewer in the eye. In research involving older people, CAWI can be exclusionary, whereas CATI or CAPI can reach people where the internet is unavailable or used less frequently.<\/p>\n
Face-to-face interviewing remains the technique that offers the greatest control over the research situation. The interviewer verifies the respondent’s identity to a limited extent, guides them through complex question blocks, and shows them respondent cards with visual materials, product concepts, packaging, and advertisements. CAPI supports long questionnaires – 40, 60, or even 90 minutes – although data quality may also decline in very long interviews. It also enables studies in locations where address-based sampling or random-route sampling are sensible ways of reaching respondents – for example, in studies representative of small towns or among segments with low internet penetration.<\/p>\n
Telephone surveys work well where speed of fieldwork, the geographic dispersion of the sample, and contact with groups that are more difficult to reach online matter – farmers, rural residents, older people, and microbusiness owners. However, CATI has a practical length limit: above 20-25 minutes, the proportion of terminated interviews increases, and respondents begin to shorten their answers. Telephone surveys are also unsuitable for research requiring the presentation of visual stimuli, such as packaging, layouts, or advertising materials.<\/p>\n
Online surveys are the fastest and least expensive technique, now dominant in many B2C studies and in many B2B studies conducted using contact databases or industry panels. CAWI enables dynamic question paths, randomization, product tests using multimedia materials, conjoint analysis, and MaxDiff. Its weaknesses include limited control over who actually responds, the risk of “clicking without reading,” self-selection among panel members, and the exclusion of people without internet access or with low digital literacy.<\/p>\n
As Hume’s Institute experts point out, selecting a method is not a matter of cost – it is a decision about whose voices will actually be heard. If a study representative of the general population is conducted exclusively online, its results may primarily describe the reality of people active online rather than that of the entire population. If it is conducted exclusively by telephone using landline numbers, it will describe the reality of households that still have a landline phone. The method itself defines the boundaries for generalizing the findings.<\/p>\n
The key dimensions in which these techniques differ operationally can be summarized as follows:<\/p>\n
The first and most common error is treating CAWI as the default method for every project because it is inexpensive and fast. In Hume’s Institute projects, studies described as “representative of Poland’s population” but conducted exclusively through an online panel often overrepresent younger, educated people living in large cities and active as consumers. Post-stratification weighting partly corrects demographic distributions, but it does not address differences in attitudes between people included and not included in the panel.<\/p>\n
The second error is ignoring the method effect in tracking studies. If the first wave of a study was conducted using CATI and the next using CAWI, part of the observed “change” results not from changes in respondents’ attitudes but from the change in technique. Mode effect can shift indicators by several percentage points – in one direction for questions about socially desirable behaviors and in the other for reported consumption of sensitive products. Any change of method in a tracking study should be preceded by parallel measurement.<\/p>\n
The third error is selecting a method that is inappropriate for the target group in B2B studies. Decision-makers in large companies rarely respond to online surveys from unknown senders and rarely answer calls from numbers not included in their contact lists. In this segment, telephone interviews scheduled in advance, invitations sent from a recognizable source, or individually recruited face-to-face interviews may achieve higher response rates than CAWI surveys distributed through cold email invitations.<\/p>\n
The fourth error is underestimating interviewer effect in CAPI and CATI. The presence of an interviewer improves respondents’ understanding of questions, but it also creates pressure to provide socially acceptable answers. In research on health habits, personal finances, attitudes toward minorities, or consumer behaviors subject to moral judgment, CAWI often produces more reliable data than face-to-face interviews despite offering less control over the sample.<\/p>\n
Combining techniques within a single project is becoming increasingly common in order to reconcile their advantages and reduce the weaknesses of individual methods. A mixed-mode design involves reaching the same population through different channels – for example, part of the sample through CAWI, part through CATI, and part through CAPI – in order to correct the biases of individual techniques.<\/p>\n
Typical scenarios in which a mixed-mode approach works well include:<\/p>\n
However, mixed-mode research requires analytical caution. Responses collected through different techniques must be comparable – which means identical question wording, the same scales, and awareness that mode effect may still influence distributions even under these conditions. A well-designed mixed-mode study includes the “technique” variable in the analysis and reports results while retaining this information.<\/p>\n
CAWI works well when the target group is well represented online – for example, users of specific digital services, customers of brands with contact databases, or B2B industry panel members. It is also an appropriate choice for research on sensitive topics, where anonymity increases the honesty of responses, and for concept tests involving visual and multimedia materials.<\/p>\n
Face-to-face interviews reach people who do not use the internet, do not answer calls from unknown numbers, or are not included in research panels. In studies representative of small towns, rural populations, seniors, or disadvantaged groups, CAPI using address-based sampling or random-route sampling can produce a structure closer to the actual population than an online-only study. It also allows for better verification of who is actually providing the answers.<\/p>\n
CATI works well for geographically dispersed target groups – farmers, microbusiness owners, and residents of sparsely populated areas. Telephone contact makes it possible to schedule an interview, increasing the effectiveness of reaching people who are busy professionally. The limitations remain the declining proportion of people who answer calls from unknown numbers and the inability to present visual stimuli.<\/p>\n