Pitfall in Cluster Sample

Cluster Technique is employed in various  consumer surveys to determine Brand preference and Brand usage surveys but all these cluster surveys fail to determine the importance of Distribution network of a particular brand having an impact in final cluster being formed. Distribution network determines the uneven penetration of a particular brand in a particular area.  Cluster sampling technique techniques has its own weakness for e.g. If the availability of Brand A is available in twice as many outlet than Brand B then it is expected that Brand A would have twice the availability as Brand B. however here Distribution of Brand A and B need not follow a consistent pattern. It may happen that the cluster points in the sample fall at random too heavily in the neighborhood of outlets for one particular brand. Especially where convenience plays an important part in consumer habits, such an unbalanced distribution of clusters can prove disastrous to attempted measurements of brand preference and usage. Because convenience is a major factor in the purchase of such varied products as ice cream, beer, and gasoline, cluster samples may be particularly misleading for these items. The problem also extends into grocery and drug items, where different chain stores push different brands, since convenience largely determines which store is patronized by people living in a given area cluster analysis could fail in this kind of scenario.This can be explained by following example which would illustrate pitfalls in Cluster Analysis.

    A survey of 450 personnel in 90 clusters was taken. The centre point of first 90 cluster and second 90 cluster was taken independently. The same sampling devices were used each time—a randomizing area selection grid applied to a detailed street map of the city. Results of the two surveys on the question on brand of gasoline usually purchased. There was a sizeable shift in percentage of people buying first two brand ,Brand B which was the Second Brand in which number of users varied from 21% in January to 15% in February while that of First brand increased from 37% in January to 42 % in February. Consideration of this drastic apparent drop in brand B's share of the market brings up the question of how the randomly selected cluster points in the two surveys may have fallen with respect to brand B stations. Plotting the station locations on the map reveals that although 28 of the 90 cluster points in the January survey were within a half mile of brand B stations, only 21 in the June survey came that close to brand B stations. Conversely, in the January survey only 44 of the cluster points had been within a half mile of brand A stations, whereas 51 of them fell that close to brand A stations in June. Changes this great (7 each way) in the number of cluster points falling near to brand A and brand B stations would normally occur in paired samples of 90 cluster points about 25 per cent of the time, so the two samples are not particularly freakish. Nevertheless, these few changes could be enough to explain the supposed shifts in brand usage.

In other words, it would be possible, under the assumption that brand usage is determined by nearness to filling stations, to reach the conclusion that there had been no shift in brand usage between the two surveys. Thus, chance variations in the locations of cluster points might be the only reason for the apparent 6 per cent drop in brand B users.

While this illustration has been based on gasoline brand usage, its implications are of wide general applicability. Whenever block samples, route samples, or other methods which tend to concentrate

interviews in groups or clusters, are being considered for consumer surveys, it is important to find out the part that convenience may play in buying habits and whether the different brands actually do have uniformly widespread availability. If convenience is a factor and outlets are spotty which is frequently the case, then the cluster sample may prove inefficient.

Anoop Menon

12126