Cross-sectional vs longitudinal survey

What a snapshot can and cannot say

In a cross sectional study, as a methodology paper in the Indian Journal of Dermatology puts it, the investigator measures the outcome and the exposures at the same time, and because it is a one time measurement it is difficult to derive causal relationships from it. A survey that finds staff who work from home report higher satisfaction cannot tell whether working from home made them happier or happier people were more likely to arrange it. That limitation does not make the design weak. Most practical questions are about the present: how many customers want evening appointments, which suburbs a new service should reach first, what share of members read the newsletter. A single well drawn survey answers those quickly and cheaply. It only fails when somebody later compares it with a different survey and calls the gap a trend.

Two kinds of longitudinal design

A trend design asks the same questions of a fresh sample from the same population each round. It shows how the population changes, but not how any individual changes, because different people answer each time. A panel design asks the same people repeatedly, which shows individual change and makes the order of events visible. SurveyMonkey's guide describes panel studies, cohort studies that follow people who share a characteristic or experience, and retrospective studies that ask about the past. Pew Research Center's American Trends Panel shows what a serious panel involves: created in 2014, about 10,000 randomly selected adults, typically one or two surveys a month, new members added each year because some stop taking part, and members from overrepresented groups retired to keep the panel balanced.

Attrition and practice effects

Panels have two characteristic weaknesses. The first is attrition. People move, lose interest or change jobs, and the ones who leave are rarely a random selection, so a panel can drift towards its most loyal members over time. SurveyMonkey's guide notes that most longitudinal studies have some churn and that a strictly longitudinal study is near impossible when employees and customers come and go. The second is conditioning: people who answer the same questions repeatedly learn them, think about the topics between rounds, and may answer differently from someone meeting the questions fresh. Refreshing a panel with new members, as Pew does, limits both problems. Trend designs avoid them, which is why many workplace and customer programmes run repeated snapshots rather than true panels, accepting that they will see change in the group but not in individuals.

Keeping a repeated survey comparable

A series only means something while the measurement stays still. Keep the question wording, the answer options, the scale length and direction, the order of questions and the time of month the same. Change any of them and the next result may move for reasons that have nothing to do with the thing being tracked. When a change is genuinely needed, make it deliberately and mark the break in every chart. In a form, publishing writes a version, and each response keeps the version number it was submitted against, so an export shows which wording every row answered. An older version can still be opened at its own address to check exactly what people saw.

Linking the same people across rounds

A trend design needs nothing extra. A panel needs a way to join a person's answers from one round to the next, and a stored form response carries no account, IP address or browser details that could do it. The link has to be a code the respondent enters each time, such as a staff number or a personal code issued with the first invitation. That choice ends anonymity, so it has to be stated honestly in the invitation. Where anonymity matters more than individual change, run a trend design instead: a fresh round each time, no code, and results read only at the level of the group. Most staff and customer programmes are better served that way.

Questions people ask