Statistical Analysis Methods

Amplitude Research offers its clients a diverse list of statistical consulting services applied by its staff of experienced statisticians and research professionals. This allows us to act as "statistical detectives" in reviewing client survey data, and applying those data analysis methods that best define the data given our client's study objectives. Some of the statistical consulting techniques commonly applied by Amplitude Research include the following:

• Statistical testing – t-tests, Chi-square, ANOVA, etc.
• Weighting data
• Correlation analysis
• CHAID analysis
• Cluster analysis, latent class analysis
• Choice Based Conjoint / CBC analysis, choice modeling
• Discriminant analysis
• Factor analysis
• Gap Analysis
• Logistic regression
• Multiple linear regression, econometric modeling
• Perceptual mapping, and many more...

When selecting records from a customer database, we often use "simple random sampling" based on random numbers assigned to the database records. However, in many cases, the database will contain information that we can use for "stratified random sampling." For example, if each record has a designation for gender, we can divide the database into two parts. Then, we can take a random sample of males; and, separately, we can take a second random sample of females. If we knew from analysis of the database that 25% of the customers were male and 75% were female, we could use "proportionate stratified random sampling," which would yield sample that is exactly 25% male and 75% female. That is, we would randomly sample three times as many females from the female part of the database as we sample males from the male part of the database.

Statistical Methods Using Disproportionate Sampling

Alternatively, there are situations in which we may want to use statistical methods with "disproportionate stratified random sampling." As one example, we might want to select the same number of sample records for males and females, if we wanted to complete the same number of surveys with males as with females. This would mean a gender mix of 50% male / 50% female for the survey, even if the customer database had a gender mix of 25% / 75%. One reason for doing this might be to obtain a sufficient sample size of customers for both genders within a particular budget limit that would allow reliable comparisons of the survey results by gender.

Then, as needed, we could engage in weighting data to properly reflect the influence of each gender while still applying other statistical analysis methods to produce a comprehensive report of key findings. For example, if in addition to comparing results between males and females, we also wanted to examine the total sample results, we could "weight" the data by gender. In this example, "weighting" would assign 75% of the influence on the total sample to the females surveyed, and 25% of the influence on the total sample to the males surveyed. When statistical data analysis is conducted by weighting data in this way, the weighting would reflect the proper gender mix of 25% male / 75% female. Even though the "unweighted" gender mix was 50% / 50%, weighting the data adjusts the results to be consistent with what would be expected for a sample with a 25% male / 75% female gender mix.