Blended Calibration
The advantages of "blended calibration"
In the case of non-probability samples, the composition of the sample is not known, which is why no inclusion probability can be determined for its members.
Therefore, no statements can be made about the standard error and – unlike with a random sample – it cannot be reduced by increasing the number of cases.
Tests show that the results of non-probability samples deviate more strongly from the true value than those of probability samples.
“Blended calibration” makes it possible to achieve an accuracy with reasonable effort that would otherwise only be possible with an extensive random sample.
Both variants are combined for this purpose: A smaller probability sample contributes to accuracy and a non-probability sample – often in the form of an online survey in the access panel – makes it possible to supplement high case numbers at low cost.
Particularly in online surveys as non-probability samples, it makes sense to use differences in communication behavior, especially in the digital sphere, as additional calibration features.
Information provided by respondents on mobile, landline and internet use and on social media activities is suitable for this purpose.
These characteristics usually vary between participants in online access panels and people selected by random sampling.
Two surveys with identical questions should be planned.
A telephone survey with an ADM sample and 1,000 cases is suitable for the probability sample, for example.
An online access panel can be used for the non-probability sample, for example from our sister company infas quo GmbH.
The number of cases is flexible and results from study-specific considerations.
In both surveys, around six additional questions are asked that are required for calibration.
In principle, the method is also suitable for a panel sample.
The exact design depends on numerous parameters (frequency, number of cases, topic) and must be clarified for each individual case.
For the initial wave in particular, “blended calibration” offers the opportunity to build up a large panel population comparatively cheaply, which can be repeatedly surveyed online.
A survey with “blended calibration” utilizes the positive characteristics of a probability sample, but does not fully match them.
The method is suitable for numerous study topics, but there are topics for which a pure probability sample is indispensable.
“Blended calibration only works satisfactorily if the non-probability sample is of high quality, i.e. if all quality assurance measures are implemented.
“Blended calibration is always an option when large numbers of cases are required.
For example, in the case of low prevalences that make screening using a probability sample very time-consuming.
Blended calibration can also be an option if fast field times are required: High case numbers require significantly more time with a probability sample than with a non-probability sample.
The practical application
infas already regularly uses “Blended Calibration” together with its sister company infas 360.
The infas panel is used as a smaller survey with a random sample.
It comprises around n=1,000 respondents per month.
The basis is a dual-frame telephone sample according to the ADM design with a mix of 70 percent landline and 30 percent mobile.
The considerably larger non-probability sample is an online survey with a net sample of n=10,000 cases.
It is based on a classic online access panel.
In addition, infas used “Blended Calibration” in the surveys for the thematic reports“Home office in the course of the corona pandemic” and“Vaccination rate and willingness to vaccinate” as well as for the mobility report “
Local public transport plays an important role in achieving climate targets and ensuring future mobility in metropolitan areas and rural regions alike. At the same time, public transport is in strong competition with private transport and new mobility services.
Practical experience to date has shown that a high-quality random sample for the reference data set is a prerequisite for the successful application of blended calibration.
The non-probability sample should also be carefully created and of high quality.
The calibration features that infas has applied and tested in various projects and in-house studies reliably achieve optimization.
Further features for additional improvements are conceivable here.
Ideally, these should be developed in the context of surveys for which there are non-empirical reference values, for example in the context of election research or in the real estate sector, where cadastral data is available.
“Blended calibration is not suitable for every challenge facing market and social research, but it is a promising building block that combines the best of two different worlds.
Additional information:
News: infas introduces a method for calibrating non-probability surveys using probability samples
