Surveys and Sampling Services Post Duran from PFC -

Statistical Sampling

Our statistics team has experience designing, implementing, and reviewing samples in a wide range of applications. Designing a proper sample is often more than simply creating a randomized list in Excel.

It may also require knowing what the sample will be used for, understanding the potential characteristics of the population being sampled, determining whether simple random samples, cluster samples, stratified samples, model based samples, convenience samples, or even purposeful samples are appropriate.

After a sample is identified, there may be need to do post-sample analysis to confirm that the implementation of the sample did not introduce bias or error, and to test various assumptions that may have been made when designing the sample. In some circumstances, it may be necessary to do additional research to assess the potential for various biases in the sample.

Once data are obtained from a sample (or multiple samples), it is often appropriate to compute descriptive statistics using classical and computationally advanced statistical techniques such as bootstrapping and resampling. Hypotheses can be tested using appropriate techniques, sometimes including parametric and nonparametric methods.

The Phillips, Fractor & Co. statistics team has provided such sampling and analytical services to government agencies, trade associations, corporations, and litigants in a wide range of industries and applications in addition to a diverse range of published academic research using sampling methods.

A properly designed sample can often provide cost-efficient estimates of population characteristics. (It is an odd feature of sampling that the required sample size for a population of 5,000 is often nearly the same as that for a population of 500,000 or 500,000,000. The sample size for most populations does not generally depend much on how large the underlying population is that is being sampled.)

Some of the common applications of sampling include selecting company records to estimate company wide totals, selecting windows in a housing development to test for proper installation, employee records to determine the estimated impact of a compensation policy, and purchasers of a particular consumer item to test failure rates.

A few more unusual applications of sampling include name recognition testing (sometimes used when there are allegations of trademark and logo violations), presence on the Internet (sometimes used when images were used without permission or in violation of contract), brand preference (sometimes used when applying marketing research methods to identify competing products or brands in antitrust and industrial organization cases), and analysis of selected zip code level data to associate disease clusters with potential causes.

And, increasingly, class action, collective action, and mass tort attorneys are finding it helpful to have a random sample of potential class members to be deposed or to sign declarations regarding possible liability and damage issues. If the sample is randomly chosen, then sometimes those data can be used for broader extrapolation to the underlying population. PFC experts are experienced at identifying a statistically appropriate number of deponents or declarants and then identifying an appropriate random sample to be queried.

Whether you are conducting an employment case, commercial liability matter, other litigation, or simply need assistance with a research project, the PFC statistics team is prepared to offer creative, scientifically appropriate, understandable sampling and statistical solutions appropriate for your data intensive problems.

Surveys, Census, and Questionnaires

There are times when it is helpful to use a survey or questionnaire to understand a population’s experience.

The Phillips, Fractor & Co. survey team conducts a wide range of surveys including Internet based, mail, telephonic, interview, and mixed-mode surveys as appropriate.

When a survey-type instrument is offered to all, or nearly all, individuals in a population, the results are like a census in that there was no intentional sampling of the underlying population; an attempt was made to get the entire population. In such cases, it may be important to attempt to analyze characteristics of non-respondents to assess potential nonresponse bias.

When a survey-type instrument is offered to a randomly selected subgroup of a population, the results are like a traditional survey and additional research may be needed to assess the representativeness of the sample and the respondents within the sample. The implementation of a survey may cost less than a full population census, but it may require additional research that would not be needed in a full census or full enumeration of the population.

When a questionnaire is offered to individuals who attend a meeting, or who voluntarily respond to an advertisement, the resulting data is usually not a random sample. The collection of individuals who provide questionnaires may have stronger opinions on some topics than the population as a whole or may have some other common feature that is not representative of the overall population. But, for class action and mass tort analysis, those who answer questionnaires at meetings may provide a minimum group for a class. In such cases, it may be appropriate to conduct additional research to compare the results of the purposeful sample (the questionnaire results) to survey results from a randomly selected subgroup or to other data sources.

The PFC survey team has created and implemented census-type research, surveys, meeting questionnaires, focus groups, and similar question oriented research for a wide range of clients and applications and is ready to discuss an appropriate solution to your survey problem with you.