7 August, 2020, Narodny Opros
Dear friends!
We all know that independent opinion polls are de facto forbidden in Belarus. We also know that online surveys carried out without due scientific rigour do not carry representative weight. Still, the citizens of Belarus want to know the real ratings of all candidates. Hence, we conducted an independent, scientifically rigorous and completely innovative Public Survey.
We used two methods: an online survey of Viber messenger users and a street survey across the nation. Afterwards, we processed these data statistically. You can familiarise yourself with these methods in a dedicated section.
With your help, we have collected almost 45,000 votes!
And here are the results:
What do they mean?
They mean that, with 95% confidence, the rating of Sviatlana Tsikhanouskaya lies between 67% and 73%, whereas Aliaksandr Lukashenka’s rating is between 16% and 21%.
Research methodology
For this research, we picked Viber as the most popular messenger in Belarus. But we were aware that the Viber poll would only cover people using smartphones and this particular messaging app, which would cause a bias towards a certain audience. Therefore we also conducted a street poll, the traditional method of polling used in sociology, in order to cover all categories of citizens and to confirm the credibility of our study. With the help of specially instructed volunteers, we surveyed over 1,000 people from all age groups and regions of Belarus.
Our survey was stratified by locality type (urban vs. rural). We surveyed groups in different communities ranging from Minsk, the capital, to small villages. We used the official data from Belstat (the governmental statistics committee) to calculate the representative size of the statistical group for each category. We are going to publish the results of the street poll soon.
In order to obtain the final candidate ratings, we took the following data processing steps:
1. Normalised the data collected on the streets and from the Viber polls using upsampling (a statistical method). The cases from the two survey result datasets were weighted to make the them comparable.
2. Used the MRP (multi-level regression with poststratification) statistical method to compute each candidate’s rating. The method involves building a model that removes the biases in the merged sample and generalizes (predicts) the poll answers to the whole country’s population.
3. The model was statistically tested and used to predict the candidates’ ratings with a 95% confidence level. For example, the model predicts Sviatlana Tsikhanouskaya’s rating to be 70±3% with 95% confidence.
Street opinion poll summary
• Over 1100 completed questionnaires.
• Representativeness: quotas by region, age, and gender.
• The selection of people to survey was random.
• Conducted by volunteers across the country from August 2nd to August 7th.
• The final calculations were carried out by reweighting the results of each questionnaire relative to the quota based the settlement type, gender, and age.
Preliminary voting:
• Did not plan to participate in early voting – 79.4%
• Planned or already participated in early voting – 14.5%
• Were not sure – 6.1%
We are open to publicising more details of the survey and justifying its objectivity. Please leave your suggestions and comments here.
The results of the street survey (for the people who were able to answer) are statistically consistent with those of the Viber survey.
We compared political preferences for different groups and organised the results by the age, education, income level, location and the preferred sources of information. The results of the study are presented on the diagrams below:
