Longitudinal weights can now be applied to any analysis of data from the first five waves of the self-completion panel survey (CRONOS-3) – fielded as part of the Infra4NextGen project. But what does that actually mean?

Each wave contains about 18-20 items for each of the five pillars of the project, which are the key areas of the NextGenerationEU recovery plan: Digital, Equal, Green, Healthy and Strong.

Users can now combine the data from the five waves and analyse items from each theme together, using these longitudinal weights.

But what are longitudinal weights? 

Longitudinal weights are statistical adjustments used when you analyse the same group of people across several survey waves. 
Cronos-3 is a panel survey, meaning that the same people are invited to answer in Wave 1, Wave 2, Wave 3, and so on. The innevitable issue is that some people will stop responding along the way, and the people who remain in all five waves are likely to be systematically different from the people who drop out. 

For example, from a group of 1,000 participants, perhaps only 650 respond consistently and remain to the very end. If it turns out that younger participants are more likely to stop participating, the remaining 650 might contain disporportinately many older people. Simply analyzing those 650 as if nothing has changed about the study group could therefore lead to a biased conclusion. 

A longitudinal weight compensates for this by attributing different weights to different respondents. In simple tems, a longitudinal weight asks: given what we know about this respondent, how likely were they to remain in the survey and respond across all five waves?

People from groups that were less likely to remain in the panel are given somewhat more statistical weight, while people from groups that were particularly likely to remain are given less. The aim is for the combined five-wave sample to better represent the population it originally came from. For example, Repondent A might count as 0.8 people as they belong to a likely overrepresented group, while Respondent B, who is far more likely to have dropped out, counts as 1.4 people. 

Applying the weights

Which weight you should use depends on whether you are analysing a single CRONOS-3 wave or combining data from several waves.

For analysis using two or more of the first five waves, researchers should use the new longitudinal weight, longweight. This includes an adjustment for differences in population size between the participating countries and should therefore also be used when analysing the integrated dataset containing all five waves.

A second version, longpspwt, excludes this population-size adjustment. This can be useful for single-country analysis or when comparing countries without giving countries with larger populations greater influence on the overall results.

The longitudinal weights are based on a statistical model estimating how likely each participant was to respond across all five waves. The model uses information collected during European Social Survey Rounds 10 and 11, when participants were originally recruited, together with information from CRONOS-3 Wave 2. Wave 2 was chosen as the baseline because it had the largest number of respondents.

If you are analysing only one CRONOS-3 wave, you should instead use the weight created specifically for that wave, such as w1weight, w2weight or w3weight. These weights correct for differences in respondents' likelihood of being included in and participating in that particular wave.

In short:

• One wave: use the relevant wxweight.
• Two or more of Waves 1–5: use longweight.
• Two or more waves without the population-size correction: use longpspwt.

Applying the appropriate weight helps ensure that estimates are as representative as possible, despite differences in who was originally selected for the survey and who continued to participate over time.

All the data is available free of charge via the ESS Data Portal

View data from all 450 questions fielded across five waves by theme: Browse by Theme