Question 1

What is statistical analysis in market research?

Accepted Answer

In market research, statistical analysis is used to identify patterns and quantify the confidence you can have in your findings. Statistics can describe a situation, such as current and past performance on KPIs, identify and quantify important relationships, such as drivers of KPI performance, and predict future events, such as how performance in KPIs is likely to change when you make measurable improvements on its drivers. According to the 2022 GRIT Insights Practice Report, 72% of research projects conducted by buyers of insights services include some quantitative data. Coincidentally, suppliers of insights services also said that 72% of their research projects include quantitative data. This means that they also use some kind of statistical analysis on most of their projects, even if they are only basic descriptives. Over the past two years, the percentage of buyers who say that investing in technology for analytics is a key priority has risen from 52% to 59%, while for suppliers the percentage has held steady in the mid-50%s. Also over the past two years, the percentage of buyers who are using or trying text analytics has risen from 53% to 66%, social media analytics has held steady in the mid-60%s, and Big Data analytics has hovered just under 60%. Clearly, statistical analysis plays a huge role in market research, not only in traditional areas, but also in the analysis of large, unstructured data sets. Common applications of statistical methods in market research include comparing performance and positioning across competitors, quantifying satisfaction and purchasing drivers, identifying new market segments, sizing market opportunities, and testing new product concepts. Statistical methods also drive marketing mix and market simulation models. With the advent of Big Data and the evolution of tools to analyze it, statistics can help predict individual behavior with greater precision and optimize a variety of customer and user experiences. As important as these applications can be, they are much less useful if you can’t be confident in the results. Every statistical method produces one or more metrics that quantify confidence in results or evaluate performance of statistical models. These vary from method to method and situation to situation, and they can also vary from statistician to statistician. Often, a statistical analysis will produce multiple measures of confidence, parsimony, goodness-of-fit, and so on, and different statisticians may value and apply them differently. A company, agency, or supplier that is expert in statistical analysis can walk you through these different metrics and help you make the best decisions about how to interpret your results.

Question 2

What are different methods of statistical analysis?

Accepted Answer

Market researchers routinely employ a variety of statistical methods, and statistical methods drive artificial intelligence and machine learning analyses. Visualization and charting are also used to analyze data, and some of these may rely on or otherwise employ statistical methods. There are three types of standard statistical methods: those which require one variable (univariate), those which require two (bivariate), and those which apply to more than two variables (multivariate). Examples of univariate analyses include frequency distributions and basic descriptives, such as means, modes, medians, and so on. It’s preferable to begin with univariate analyses because it increases your familiarity with the data and helps to identify potential data quality issues. Bivariate analyses, such as cross-tabulations or correlations, increase your understanding of the data set, especially how one variable may be related to another. It is a useful step to take in order to develop hypotheses to guide the next phases of your analysis. Once you understand how your data is structured and developed some working hypotheses, you can proceed with multivariate analysis to identify more powerful insights. Of these methods, regression analysis is probably the most commonly known and generally provides the foundation for the other methods. Regression analysis determines how strongly each one of multiple variables (the independent variables) are related to another variable (the “dependent” variable, so named because its value is hypothesized to depend on the values of the other variables). A common application of regression analysis is quantifying the drivers of customer satisfaction. Another multivariate method, discriminant analysis, can be thought of as having multiple dependent variables in addition to multiple independent ones. It attempts to quantify how well each of a set of factors distinguishes between other variables. An example might be understanding market perceptions of competitive positioning. If each competitor is rated on the same set of factors, discriminant analysis will tell you how well each factor identifies each competitor. If there are three competitors, and one achieves high ratings on a factor, one rates in the middle, and one rates near the bottom, that factor is highly discriminating. If all competitors have similar scores on another factor, that factor does not discriminate very well. The analysis helps you to understand competitive strengths and weakness as well as how each factor is related to the others. Two other common multivariate methods, factor and cluster analysis, do not have dependent variables. Factor analysis is a data reduction tool that looks at similarities and differences across variables and groups them into a set of “factors.” An application of factor analysis might be to simplify complex sets of attitudes. Suppose you have asked for ratings across a battery of 15 attitude statements. You can use factor analysis to reduce these to a handful of concepts that are easier to comprehend. Cluster analysis groups respondents or cases together based on similarities across a set of variables. Its goal is to maximize similarity within a cluster while maximizing differences across clusters; a common application is segmentation. Many more complex statistical methods are also available. For example, regression, factor, and cluster analysis can also be executed via an approach called latent class modeling. This methodology assumes that the observed, or measured, variables are related to unobserved, or latent variables, and this enables it to model the data. This has a side benefit of producing a wider array of metrics that can be used to evaluate the result. Another set of modeling approaches focuses on understanding how respondents prioritize, or trade off, various criteria when making a decision. These methods include choice modeling or choice-based conjoint and maximum difference scaling, or maxdiff. These produce relative values for a set of decision alternatives and are sometimes used to create preference simulation models. Especially in relation to choice modeling but applicable to other methods as well, you may hear the term “hierarchical Bayes” or “HB.” At the risk of oversimplification, HB models estimate parameters by guessing the value for each respondent based on the answers provided by the entire sample, then adjusting it based on how that respondent actually answered. By leveraging the robustness of the full sample, you can make better estimates at the individual or segment level where data is sparser. Statistical concepts also drive artificial intelligence and machine learning analyses. More specific examples of these methods include neural networks, decision trees, natural language processing (NLP), and evolutionary programming.

Question 3

What kinds of data can be analyzed with statistics?

Accepted Answer

Statistical analysis can be applied to many types of data, both structured, such as survey data, and unstructured, such as web usage. You can think of structured data as data that is collected according to an experimental design which dictates which people or data points are valid to analyze and the scope of the data required. It’s generated by an artificial process for the specific purpose of researching a defined issue or set of issues. Unstructured data can also be used to research a defined issue or set of issues, but that is not the primary reason why it exists. Location data and web traffic data, for example, exist because people generate it when they move from place to place, and service providers use it to provide context-sensitive assistance to them. These data can be analyzed for other purposes once collected, but there is no design that dictates who can provide data and which locations or websites they can visit. While both structured and unstructured data must be subject to rigorous QA/QC procedures, structured data should require less effort because some level of quality and quality flags have been designed into it from the start. Unstructured qualitative data, such as text from blogs, focus groups, or social media posts, can also be analyzed via statistics, but only after you transform it into quantitative data. This can mean manually developing a code frame and making judgments about which codes apply to each word or phrase in the data set. Some research suppliers may also offer tools that can assist in developing coding rules, minimizing the pain of the manual work. The use of structured data in market research is as old as market research itself, and the ever-increasing amount and accessibility of unstructured data and tools to analyze it have greatly expanded the role of statistical analysis to support all areas of marketing, advertising, and product and service development.

Question 4

Why do companies hire external experts in statistical analysis?

Accepted Answer

Some common reasons why insights professionals hire experts in statistical analysis are to access capabilities they lack and to increase their capacity for analysis, but perhaps the most important reason is to ensure objectivity. Objectivity is needed during every step from research design and data collection through making recommendations, and external experts are not as likely to be influenced by biases as members of your internal staff. To say that someone is biased is not to accuse them of being dishonest, only of being human. We all have biases and they can manifest themselves in different ways. In this case, one obvious example of bias is extent of expertise. An external statistical analyst may see opportunities to add value via methods that your team does not know about and could not apply. They might also look at your universe of prospects differently. Sometimes, when someone is mandated to be “customer-focused,” they study their own customers and neglect to study potential or lapsed customers. An external expert might not make a mistake like this. An external expert may also add value by increasing the accuracy of your research and findings. First of all, they may be more likely to identify holes in your project design because someone who is responsible for analysis also has to take responsibility for the data whether they want to or not. In the Supplier Performance section of the 2022 GRIT Insights Practice Report, we hypothesize that some data and analytics suppliers may be hired into no-win situations because they were brought into the project after the (bad) data has been collected, and there is no way to recover from that. They should be especially sensitive to these situations, and therefore very helpful. Another way in which they increase accuracy is by being, well, experts. You may have someone on staff who can execute the same kind of statistical analysis as the external analyst, but they may have other responsibilities as well. The external statistical analyst is more likely to spend more of their time in this area and be less likely to make mistakes or poor decisions. They will understand which methods to use, which metrics to follow, and which patterns or anomalies are red flags. The external expert may also be more willing to “draw outside the lines” than someone who simply took a course. The popular conception might be that statistics are science, and science is bound by strict laws. The reality is that statistics deal with the unknown and probabilities and following a “cookbook” hampers your ability to find the best answer. An external expert is able to make those judgment calls and convince stakeholders they were the right ones to make. According the 2022 GRIT Insights Practice Report, over 70% of buyers of insights services use an external data and analytics supplier at least occasionally, and nearly one-third use one regularly. Also, 70% say that one of their significant roles is “data analyst,” meaning that there is significant overlap between those acting as data analysts and those hiring external experts. This leads us to hypothesize that external suppliers of statistical analysis might be augmenting capacity, but they are also likely to be complementing expertise. Given the prominent use of text, social media, and Big Data analytics, it seems reasonable to hypothesize that external statistical analysis suppliers are a key resource for these methodologies. Further, “democratization of data” is a bit like “democratization of heart surgery” – maybe the amateur will get it right, but will you bet your life on it? People who advocate for democratization of data seem to forget that democracies cannot last unless they are preceded by literacy, and data literacy would be an important foundation for its democratization. Experts in statistical analysis can help you become more literate or just make the right decisions.

Question 5

What are the 4 most common misconceptions about statistical analysis?

Accepted Answer

Four misconceptions about statistical analysis seem to persist, such as the belief that the data is more objective than internal experts, the assertion that “the data should decide,” data can be used “as is,” and that there is an absolute meaning to “statistical significance.” External experts in statistical analysis can guide you through these. First, the data could be more objective than internal experts, but that depends on how objectively the project was designed, how well the quality of the data was assessed, and how objective the data analysts were. Internal experts, if they really are experts, have deep experience. Others may consider it to be anecdotal, but they have a lot of anecdotes plus they have “wisdom.” It’s probably smart to engage them in the design of the research because they can add a perspective that others lack. Project work plans and analysis plans don’t write themselves, and the person writing them needs a broad perspective plus good judgment and willingness to take a position. Sometimes you’ll hear someone say, “we’ll let the data decide.” The data decides nothing. It’s data, not information, insights, or recommendations. The data is one perspective, and that perspective doesn’t care too much about how your business fares. You need to understand the potential and limitations of the data plus the considerations not accounted for by that data. An external expert in statistical analysis should be able to help you keep the proper perspective. Very often, you will see people tracking live results on a survey platform, viewing the data “as is,” and getting exciting about the implications. Yet, the data has not been cleaned of bogus or incompetent respondents, and the relationships between variables has not been explored. You might see that 40% of respondents need attribute X, but what if 60% of women need attribute X? What if 90% of rural women need attribute X but only 30% of urban women need it? What if 40% of respondents need attribute X, but only if attribute Y is also provided? What if 40% of respondents need attribute X, but 90% of those are customers of a competitor who would not buy from you in 1,000,000 years? You need to do something to the data before you can use it effectively. You don’t analyze responses the way you ask the questions. Sometimes, someone – usually an executive – will demand that the research be designed to 95% confidence. Well, if you are conducting a traditional political poll, you’re in good shape. If you are doing something with an iota of sophistication, you need to take dancing lessons. The significance will depend on the how the data is structured and how the responses turn out. The important concept is that statistics deal with probabilities, not significance. Significance is your call. If you can launch only one of two alternatives, and one has an 80% chance of hitting your target and one has a 60% chance, you are out of luck if you need 95% confidence but still need to launch a product.

Top Statistical Analysis Market Research Companies

What are statistical analysis market research companies?

Featured Experts in Top Statistical Analysis Market Research Companies

Compare Statistical Analysis Companies

NewtonX

NewtonX

SIS International Research

SIS International Research

Ascendant Consulting Firm

Ascendant Consulting Firm

Mindspot Research

Mindspot Research