SRM Special Issue: Inference from Non Probability Samples


Nick Allum (University of Essex, Colchester, UK)

Ulrich Kohler (University of Potsdam, Germany)

Laurent Lesnard (Sciences Po, Paris, France)

Submission due

Articles due by September 31, 2017


In survey research, the data collected with a survey are the facts we know, from which we want to infer certain characteristics of a defined population (descriptive inference) or about a parameter of a more general data generating process (causal inference). In probabilty samples, the process that creates the observed data from the population is well known, making descriptive inferential statements from the observed data to the population relatively unproblematic. While causal statements require further assumptions, probabiliy samples ease the inference from respondents to (classes of) human beings in general subject to assumptions being met. Probability samples have thus always been an important criterion for the quality of empirical research with surveys.

However, the role of probabilty samples for survey research have come under pressure from various directions. Coverage errors and unit nonresponse place a question mark on the assumption that a probabilty sample by design remains to be a probability sample in practice. In recent years, economics witnessed an experimental revolution that brought an increased use of experimental designs with highly selective research units away from the field and into computer labs, and this experimental revolution has also reached other disciplines of the social sciences. The technical innovation of web surveys has made the design of survey experiments very easy in practice although the classic distinction between internal and external validity is still a factor in judging the worth of such experiments for robust causal inference. While the response rates in classic survey modes diminish, self-selected samples of respondents are increasingly available via social networking and specialized internet platforms.

The purpose of the special issue is to discuss the possibilities of drawing inferences from social science data without the use of probability samples, or from probability samples with low rates of response. What kind of research questions can be answered using self-selected samples, and what measures can be taken to combat selection bias? We seek submissions that criticize the use of non-probability samples and small-scale experiments, as well as submissions that advocate their use. Best practice examples are welcome, as well as methodological discussions.

Also see

The presentations of a conference on the same topic organized by the editors of the planned Special Issue can be found on Note however that this call for papers is open to everybody from everywhere, not just for those who presented on the conference.


To submit, go to the SRM website and upload the article as a PDF, just like a standard SRM article; be sure to mention the special issue in the field for Author comments

Reviewing policies losely follow the usual standards of SRM: Each submitted paper will be assigned to one of the three editors of the special issue. The supervising editor will then select two expert reviewers. Those reviews and the reading of the supervising editor will be used for the final decision.

Researchers who want to submit are requested to adhere to SRM's implementation of the Transparency and Openness Promotion guidelines of the Center for Open Science. See the Author Guidelines for a detailed description of those policies.