Statistical Consulting Services

Amplitude Research has extensive experience designing and conducting complex research studies, providing statistical data analysis, applying advanced statistical analysis methods, engaging in statistical sampling for surveys of customers and markets, examining non-response bias, and concisely summarizing study findings for decision-makers. Below are some examples to provide a brief overview of our statistical consulting capabilities:

Amplitude Research has extensive experience in statistical data analysis and reporting on data from a variety of sources, including telephone surveys, mail surveys, web surveys, and existing data sources such as customer databases, employee databases and sales / transaction records. Types of studies we often conduct which encompass statistical consulting, statistical sampling and/or statistical data analysis services include the following:

• Advertising / marketing effectiveness
• Brand awareness, image, attitude, and usage
• Concept testing, opportunity assessment
• Conjoint / CBC analysis, choice modeling
• Customer / client satisfaction surveys and loyalty measurement
• Employee satisfaction, compensation, job characteristics, etc.
• Market research surveys of all kinds
• "Lost customer" studies
• Pricing sensitivity, price elasticity
• Product or service evaluation, feature / attribute importance
• Segmentation
• Target market profiling, and many more...

Some Practical Considerations For Statistical Sampling

When considering a new research survey, one of the first steps in statistical sampling is to define the "target population," which consists of all of the individuals (or households or other types of units) relevant to the study. For example, if we were considering a customer satisfaction survey for a company, the target population would consist of the company's customers. Often, in this case, we make the definition more specific and focus only on those who have purchased a product / service from the company within a particular recent time frame.

After the target population has been defined, the second step in statistical sampling is to decide how best to sample from this population. The ideal solution is to have random sampling taken from the entire target population, but there are many considerations that impact how statistical sampling is carried out in practice. One important consideration is how to identify and contact the individuals who are members of the target population, and this can be more difficult in some situations than others, depending on the quality and type of records. For example, if e-mail addresses are available for only a small proportion of customers, then customer telephone numbers (if available) can be considered for a telephone survey, or regular mailing addresses (if available) can be considered for mail surveys.

Another step when considering a new research study and statistical sampling is to decide how many survey respondents will be needed. This is often referred to as determining the survey "sample size." Many researchers consider a statistical sampling error of +/- 5.0 percentage points to be a reasonable starting point for many research objectives, although a lower margin of error (and more survey completes) may be needed depending on the desired statistical data analysis. For a "large" target population, a sample size of 400 respondents is a common choice for statistical sampling, as this provides a margin of sampling error of +/- 4.9 percentage points (at the 95% confidence level). A smaller sample size is needed when the target population is sufficiently "small." For example, if a company had 1,400 customers in their entire customer population, then a statistical sampling of 300 respondents would be needed to achieve a margin of sampling error of +/- 5.0 percentage points. (Note that close to 400 respondents was needed for statistical sampling when the target population was large to achieve roughly the same margin of error as 300 respondents randomly selected from a small target population of 1,400 customers.)

Statistical Consulting and Non-Response Bias

Although we can take a random sample from a target population, not all of those selected for participation in the survey may be reachable and/or will agree to participate. Unfortunately, those who cannot be reached or persuaded to participate in a survey may differ (on average) from those included in the survey, and the extent of this potential problem is often referred to as "non-response bias." We usually do not have a way of gauging the full extent of non-response bias; since, by its very nature, this problem is based on the inability to collect information from non-responders.

Researchers will often try to gauge the likely extent of the problem by comparing results from the earliest responders to results from those who respond later (or only after multiple attempts or reminders). This can help to some extent, but there is still uncertainty about whether the non-responders may differ significantly from even the late responders. A statistical consulting approach sometimes used to mitigate non-response bias is to examine the response rate for particular pre-identified subgroups. If statistical data analysis reveals that there are significant differences in response rates by subgroup, then sample balancing (i.e., "weighting the data") can be used to ensure the subgroups are more properly represented in the final "weighted" analysis.

Given the uncertainty about non-response bias, some research objectives may call for taking steps to minimize non-response bias. There are many tactics that can be used in statistical data analysis, and the best approach depends on the specific research objectives, the available contact information, and the research budget for statistical consulting services. In some circumstances, a dual-mode research methodology is a good solution.