
Part 2. MM4XL Tools > 2. Analytical Tools > Sample Manager > Technicalities Sample ManagerTechnicalitiesThe sample size impacts highly on the overall cost of survey studies. If, when planning a survey, one starts from the budget this is not a good sign. Fortunately, Sample Manager can help finding acceptable compromises between accuracy and cost. Let see how. The software computes sample sizes with the common bernoullian formula:
Where:
The population of the study must be precisely identified. 18700 nurses in Texas rather than 250000 golf players in Europe or 19 millions households in Italy. Above 1 million units, populations are commonly treated as infinite, for results do not change significantly. Below 500 units it is not recommended to employ the formula above. Samples can be designed in many ways. In marketing research, the most common sampling techniques are random and quota. Quota sampling is a nonprobabilistic method applied when the population cannot be described in detail. Random samples assign to each member of the population the same probability of being selected. Every nurse of the 18700 in Texas has a probability equal to 0.00534% of being extracted randomly. And all 18700 nurses could be part of a quota sample made of other groups such as 6732 doctors, 1059 surgeons, and 3997 clerks. Samples are drawn so to reproduce certain characteristics of the population they are drawn from. The more important the answer we seek, the closer the sample should reproduce those characteristics under the probabilistic assumptions. If we were to launch a new product we may desire accurate data, say with 95% Confidence level. This means, if we draw 100 times the same sample, under the same conditions, 95 samples show very similar values. The Confidence level of survey studies for the business environment is usually set at 95%, but it is not unusual the case of levels ranging 90% to 99%. The higher the confidence level the larger the sample. Tip: The Error level is the range within which values from a survey are allowed to wave. If we found that 53% or responders answered yes to a certain question and we operated with a 2% error level, the value is interpreted as 51% <= yes <= 55%. When the range gets wide one should pay attention interpreting results. Read the chapter Proportion Analyst in this help file for a comprehensive description on significance of proportions obtained from sample surveys. The fourth value Sample Manager requires as input is the Hypothesis of the study and it is an important one because it can help reduce significantly the cost of studies without impacting on accuracy. In absence of information, the hypothesis level is typically set at 0.5, which implies the largest sample of its kind, yet samples get smaller when moving away from this base value. For instance, say we were to launch an editorial product by means of direct marketing using 1 million names from a list broker we trust. From previous experience we know the list could return some 200000 prospects to contact us. In order to test the concept feasibility we opt for an exploratory survey at the 95% confidence level, 1% error level, and 0.2 for hypothesis of the study (20% of prospects contact us). This yields a sample of 6109 individuals, which indeed would have been 9512 interviews if the hypothesis level was set at 0.5, due to lack of supplementary information. This is what we meant with Sample Manager can help finding acceptable compromises between accuracy and cost. Comparing various levels of significance and error together with robust hypothesis can help designing very convenient samples still able to supply the relevant informative content we seek. Finally, it must be remembered sampling in business is not exclusive practice of marketing research. Also sales analysis, promotional campaigns, direct marketing actions, and more can profit from sampling techniques. 