A construct is simply some variable considered at the conceptual level of analysis. Thus, every questionnaire and/or survey is used to measure a construct whether behavioral or otherwise. Consider a single-item survey with a forced-choice response:

Question: Do you currently smoke any tobacco products?

A) Yes

B) No

The construct is 'tobacco smoking behavior'. Does the question seem to "get at" the construct of interest? Yes. Face validity is established. However, we could speculate that smoking behavior entails more than simply smoking or not. In other words, this behavior may be more complex than we can measure with this single-item survey. So, the content validity of the survey could be disputed. However, one also has to consider the scope of the study and the goal of the researcher. Supposing the only aspect of smoking behavior one is interested in for a particular study is whether a participant is a current smoker or not, this single item is sufficient. If we wish to infer anything beyond that, we need a better measure. In other words, it's not just about the survey/questionnaire itself, but also what one intends to infer from the observed data.

Just my two cents

Taking your example "do you usually do X?"

-What is 'usual'? Person A's interpretation of usual could be 'daily', whereas Person B's could be a 'a few times a week'. Person C's interpretation of usual could be 'at least once a month'. We could dispute the face validity of this item due to this ambiguity. We could improve the validity of the measure by rephrasing the item and changing the response format:

Question: How often do you do X?

A) never

B) 1-2 times per week

C) 3-5 times per week

.

.

.

or use an open response format.

How many times do you do X each week? _______________________

The same is true of "is Y a standard procedure in your country?"

-Whether something is judged as standard procedure is highly subjective across individuals. One person's experience with the Y could vary considerably from someone else.

Methodologically speaking, the "descriptive" nature of a survey allows it to be used only for exploratory purposes. The poor clinical value of a survey makes it a "shallow" tool from the perspective of psychometric integrity. For instance, in constructing a survey it is unlikely to measure indexes such as concurrent or divergent validities compared to what is usually done for questionnaires, scales, or tests. Still, confirmatory factorial analyses are not usually performed, as it is done when one intends to validate an instrument. It is unusual for a survey to be validated since the validation implies standardization, and therefore generalization. Indeed, a survey is constructed in order to explore the characteristics of a phenomenon within a given population; hardly the same survey might be used to explore the same phenomenon in other populations, hence the meaninglessness of a validation. However, it is common practice to assess (i) a survey's content validity by measuring agreement between two or more experts on the representativity of the items used (e.g., Cohen's k or ICC), (ii) internal consistency by using Cronbach's alpha and Pearson's corrected item-total correlation, and (iii) to conduct, if the construct is quite multidimensional or if the survey explores more constructs, an exploratory factorial analysis (e.g., PCA). As for the face validity, note that this is a "pseudo validity". Indeed, by definition, it is the degree to which an instrument appears to measure a given construct to the layperson.

PS. in answering to your initial statement, both a survey and a questionnaire may explore one or more constructs.

Ciro Rosario Ilardi I'm a bit confused by your characterization of surveys, questionnaires, and scales. A survey is a general term for a self-report measure, which usually contains a questionnaire (the actual set of questions themselves). A scale is an emergent property of the specific group of items and the properties of the items that connect them together into a whole.

Blaine Tomkins I agree. A questionnaire is a set of questions that can be delivered in the form of a survey.

However, when I was talking about "questionnaires, scales, or tests", I was talking about tools commonly used for clinical purposes. When talking about "validating" a clinical tool (de novo or when an instrument is "imported" into another cultural context, e.g, from USA to Italy), there are different steps to follow (e.g., collection of normative data, confirmation of the factorial solution hypothesized, measurement of criterion and construct validities, and so on). These steps are clearly performed to make the instrument "valid", i.e., able to properly measure what it should measure, and "calibrated" in a specific population.

"Validating" a survey is, on the contrary, a quite ambiguous expression. On the one hand, a survey may contain clinical inventories (e.g. the Beck's Depression Inventory, the STAI-Y, etc), but these should have already been validated. On the other hand, the survey is typically designed for purely exploratory and descriptive purposes. Consequently, nobody collects data from a large normative population to validate a survey (which may include questionnaires already validated), but rather uses the survey to query the population with respect to the topic of interest. For instance, if a survey is conducted to explore, in the clinician population, the management aspects of a disease, it is extremely unusual for the survey to be first validated and then disseminated. Certainly, it is important that the survey have reasonable content validity, with high agreement among judges regarding item representativeness. Any other conclusions about the psychometric integrity of the survey would need to be drawn after collecting data from a large population, different (but similar for characteristics) from the one that will be examined later; which would make very little sense for a non-clinical instrument.

Ciro Rosario Ilardi I fear we may be using the term 'survey' in different ways. Perhaps in a clinical context, the term means something else I'm not sure. However, regardless of whether a self-report measure is being used for purely descriptive/exploratory purposes or not - validity must be considered. Instrument validation doesn't begin only after data have been collected I guess is my point. Yes, one is pretty much limited to face/content validity prior to the collection of data, but the question is asking specifically about these forms of validity. As I alluded to in the examples above, even modifying the question format or response scale can change the type and quality of information you receive from respondents.

Blaine Tomkins "even modifying the question format or response scale can change the type and quality of information you receive from respondents." Totally agree.

Don't get me wrong, statistical validation should always supersede subjective validity. When discussing the topic of 'validity', my students often ask, "Why would one even bother with face validity when there are much more informative and objective methods?"

I usually answer along the lines of "You're right. Face and content validity are far from ideal. However, when we are creating a brand new measure, we have to start somewhere before we even have data to look at". At the very least, we can take advantage of theories and evidence accumulated from the study of self-report behavior to help inform our items and avoid the creation of bad items.