If business to business marketers experience twinges of inadequacy when compared to their counterparts in consumer fields, sampling is often the reason.
Consumer marketing men trot out jargon about quotas, stratification, probabilities and the like which is mumbo jumbo to 99 per cent of industrial marketers. The business-to-business researcher feels intuitively that the sophisticated sampling methods are not appropriate. This article presents a number of simple rules which guide industrial sampling.
Before a sampling method can be decided, a certain amount of information is required on the market more correctly referred to as the "universe" or "population". The researcher should ask two very simple questions about this universe.
It is, of course, possible that a certain amount of homework is needed to answer these questions. However, only a broad view is required. At this stage it is not necessary, for example, to know who the big buyers are, only to know that some do or do not exist. Let us now discuss each type of sample in turn.
Typical of such universes are buyers of computers, household cable, copper pipe, fork lift trucks, commercial vehicles, fertilisers and bricks. This is not to say that all the companies buying these products are small, on the contrary, many may be large concerns. However, there are none that are so large that they account for more than one or two per cent of all purchases. Where business to business universes are homogeneous in this way they are similar in construction to consumer markets and consumer type sampling methods can be employed. This has great advantages as it means that samples can be selected randomly and the error in the results can be calculated mathematically.
Of course, it is necessary to have some knowledge on the construction of a universe before one can make the judgement that it is akin to the mass populations typical of the consumer world. It may be necessary to obtain this feel for a market from people with an informed view or to carry out a pilot project to test the uniformity of the universe before the sampling method is finally decided. As a general rule, however, industrial universes which correspond closely to consumer universes in structure tend to have one or more of the following characteristics.
The benefit to market researchers of homogeneous universes is the ability to select random samples and hence measure statistical error. There are, for example, between 4,000 and 5,000 electrical contractors in the UK and a random sample of 500 showed that 20 per cent are prepared to buy imported cable. We can calculate that the true proportion (i.e. it may not be exactly 20 per cent as sampling introduces errors) will be between + 3.5 per cent at 95 per cent confidence limits. That is, we can say that there is a 95 per cent chance that the true proportion of contractors that would buy imported cable will fall somewhere in the 26 range 16.5 per cent to 23.5 per cent. (The confidence limits in market research are conventionally expressed of the 95 per cent level though they could be given at 90 per cent in which case they would be narrower or 99 per cent, in which case they would be wider.) It is important to understand that the accuracy of any findings to a question depends on the proportion that answers it. If from the same sample of 500 electrical contractors, five per cent said that they currently buy imported cable, then the range of error is + 1.8 per cent (not + 3.5 per cent). That is, there is a 95 per cent chance that the true proportion lies between 3.2 per cent and 6.8 per cent.
A sample size of only 100 would produce an error of nearly + 10 per cent while a sample of 500 would give an error of just over + 4 per cent on a 50 per cent proportion. The sample size must be selected, therefore, by deciding what level of accuracy is required. In practice most researchers prefer to work with sample sizes of around 200 (or more) as on most findings this gives an error of six or seven per cent.
Random sampling can be expensive and it requires the availability of a comprehensive list of companies from which to make the selection. As a result researchers may choose to contain costs by using multi stage sampling. Multi stage sampling starts with a list of companies which is separated into groups because they have a characteristic in common. This grouping is known as a stratum. For example, if a survey is to be carried out of the replacement industrial filter cartridge market it would be necessary to draw up lists of strata of end users - ship operators, aviation fuel distributors, oil refineries, etc - all the sectors that use filter cartridges. A random sample would then be selected from each group or stratum. Equally, companies could be stratified according to size (number of employees is a simple and easily available measure) or geography.
The same strata of companies can be used as the basis for selecting a "quota" sample. As the term suggests, the researcher selects a quota or proportion of companies from each stratum in which interviews are to be carried out. Say the objective of the survey is to compare the image of a company among small, medium and large buyers of a product, lists would be drawn up, perhaps using number of employees if a correlation existed between this and buying power) and a quota of 100 or 200 (whatever) companies selected from each stratum. The quota in this case would ensure that the same numbers existed of small, medium and large companies to facilitate a comparison between the three cells.
For those industrial companies that sell their products to a finite universe of buyers of widely differing sizes, a random sample would not work. Imagine a random survey of the 200 diverse fastener industry which missed the Haden MacLellan group. The omission would very seriously skew the results. Most industrial markets follow Pareto principle where 20 per cent of firms account for 80 per cent of the volume consumed. In such markets it is vital to stratify the sample according to the size of company measured in terms of their purchasing power so that all the large buyers (there may only be half a dozen) can be interviewed while a sample is taken from the rest.
Stratifying the sample is half the problem: the researcher may not know at the beginning of a survey who is the largest and who is the smallest. In practice a bit of desk research studying ads in trade journals, entries in directories and articles in the press, should enable the list to be crudely sorted into "possibly large" and "possibly small". Given a relatively small universe of 100-200 companies, the researcher can begin to fine down the list by selecting an initial sample from the "possibly large" and asking each respondent, at the end of the interview, who else is likely to be a large buyer and worth contacting. This system of referral works well as companies in any market know their competitors.
Frequently a two stage sampling of smaller establishments is required if these run into hundreds. An extensive sample can be covered by a brief telephone interview to obtain key characteristics, thus enabling a sub sample to be drawn up for a study of greater depth - perhaps by face to face interviews.
Whether or not referral is used, the deliberate selection of companies for whatever reason, means that the sample is not random and measures of statistical accuracy cannot be applied - it has become a "judgement" sample. Judgement samples may actually be more accurate than random samples as the selection of companies could take in all the largest (thus carrying out a census of 80 per cent of the purchasers) plus a certain coverage of those of a smaller size - these having first been sorted out into strata according to the best knowledge of the researcher.
Companies commissioning research very often feel insecure with industrial samples as it is difficult (except with large homogeneous universes) to know the accuracy of the findings. Though their concern is understandable, it is also only partly germane as accuracy can also be markedly influenced by the quality of interviewing and the interpretation of results. The final abiter must be whether the results pass the common sense test - they seem reasonable in the light of other information which is known about the market.