Despite the highly developed procedures that are used for public opinion polling, the results are strongly influenced by the researcher. To the extent that the surveyor can be independent from the client, the results are less biased. This had originally been shown in a study by Hennessy and Hennessy (1961). Shamir’s paper provides additional evidence on this issue. He examined 29 Israeli polls in 1981 and 1984. The degree of independence ranged from low (in-house pollster), to medium (commissioned), to high (self supporting). The following table, adapted from Shamir, reports on the absolute deviation of a given poll from the average predicted by a group of polls. In calculating this deviation, Shamir examined the predicted difference in the vote for the two largest parties. He then compared the difference predicted by each poll, with the average difference predicted by all of the polls. For example, the table shows that 10 of the polling agencies that were independent provided forecasts close to the group mean (deviation of 2.2 or less), while none of the independents had deviations larger than 4.2. In contrast, three of the five pollsters that were low on independence differed significantly from the group mean. The impact of pollster independence was significant at p < 0.05.
The average deviation for the in-house polls was much larger than that of the independent polls. In conclusion, political parties get favorable polling results from pollsters that they hire. Might this also, occur in other countries . . . and in other organizations?