• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
Important announcements 1

News

Demographic, Contextual, and Attitudinal Factors of Attrition in Online Panels

On April, 25th at the 13th LCSR International Workshop Boris Sokolov (LCSR, Russia) presented the study "Demographic, Contextual, and Attitudinal Factors of Attrition in Online Panels".

Demographic, Contextual, and Attitudinal Factors of Attrition in Online Panels

In his presentation titled "Demographic, Contextual, and Attitudinal Factors of Attrition in Online Panels", Boris Sokolov discussed the factors that predict participants' continued participation in longitudinal online surveys. The study co-authored by Violetta Korsunova, a researcher at LCSR, and Sergei Semenov, a third-year undergraduate student in the Sociology and Social Informatics program at HSE, used data from four waves of the Russian part of the international project "Values in Crisis".

According to the results presented by Boris, men, older individuals, respondents with middle and high incomes, higher education, living in rural areas, married, and with children are more likely to remain in the online panel. Young people are most likely to drop out: among those who participated in all four waves, there were 3.5 times fewer representatives of the 18-24 age group and more than 1.5 times fewer representatives of the 25-34 age group than in the original group. In addition to demographic factors, the likelihood of continued participation is associated with higher scores on basic values (measured according to S. Schwartz's approach) of stimulation, hedonism, achievement, and universalism, and with lower scores on values of security, conformity, and tradition, as well as the personality trait of neuroticism.

However, it is unclear which factors – demographic or personal – are more likely to contribute to respondents remaining in the study: different machine learning methods (logistic LASSO regression and random forest), used to predict continued participation, produce different and almost non-overlapping sets of important variables and at the same time have a rather low accuracy. In the future, the authors plan to focus on building a more accurate model for predicting respondent retention.

The text was prepared by Boris Sokolov and Alina Moroz.