Market Research

Market Research — questions and answers

How is price sensitivity measured without simply asking what someone would pay?

Direct questions produce unreliable answers, so structured techniques are used instead. A van Westendorp series asks at what levels an item seems too cheap to trust, a bargain, getting expensive, and too expensive, then reads the crossing points. Conjoint approaches present bundles of features at different values and infer what each attribute is worth from the choices made, which behaves more like a real trade-off.

How many responses does a survey actually need?

It depends on the precision required and on how the results will be split. A few hundred responses gives a margin of error around five points for a whole-sample figure, but analysing five subgroups means each needs its own adequate count. Precision improves with the square root of the sample, so cutting the error in half takes roughly four times as many responses.

What is sampling bias, and how does it survive a large sample?

Bias comes from who is reachable and who chooses to answer, not from how many respond. A hundred thousand responses from one channel's most engaged followers still describes that group. Volume narrows random error while leaving systematic skew untouched, which is why a large convenience sample can be less trustworthy than a small carefully constructed one.

When is qualitative work the right choice rather than a survey?

When the question is why, or when the useful answers are not yet known well enough to write options for. Interviews and observation surface language, motivations and unanticipated obstacles, which then inform a survey worth fielding. Running a quantitative study first often measures the wrong things precisely, because the answer categories came from assumptions rather than from listening.

Why does a screener matter as much as the questionnaire?

It decides who enters the study, and therefore what the results describe. A weak screener admits people outside the population of interest, diluting findings, while an over-tight one produces a sample too narrow to generalise from. Screeners also catch inattentive responding, which otherwise adds noise that looks like real variation in the data.

What are the signs that a research finding is being overread?

A percentage quoted without a sample size, a subgroup difference presented without the count behind it, a causal claim from observational data, and stated purchase intent treated as forecast demand. Directional evidence used to rank options is reasonable; the same evidence used to predict a specific volume is not, and the distinction usually disappears somewhere between the report and the summary slide.