Categorizing Users 1 - Market Segmentation

Market research has been practiced for four years, and only now have I begun to think seriously about the similarities and differences between market segmentation and user personas. Been practicing when I'm busy, reading and reflecting when I'm not. Addressing this before the next practice.

I. Definition

The concept of market segmentation was first proposed in 1956 by Smith, an American marketer, "The basis of market segmentation is based on the development of the demand side of the market and for products and marketing activities to do a more reasonable and indeed adjusted to the needs of the consumer or applicant . " Alfred (1981) defines market segmentation as "the differentiation of a market into customer groups so that each group can be the target market for a particular marketing mix." According to James, author of Market Segmentation and Positioning, "the objective is to identify subgroups of customers or potential customers who will respond similarly to a given marketing mix (products/services, prices, promotions, and distribution) or who, in turn, will contribute favorably to the firm's marketing plan. In other words, it is about identifying customers who have a desire to buy a particular type of product or service that is significantly different from those in other market segments."

Second, the purpose

From the definition of scholars in different years, market segmentation are based on different user needs, serving the enterprise's products or marketing mix.

Third, the assessment of whether the segmentation is effective conditions

Theoretically, successful market segmentation must meet three conditions: 1, these subgroups have different needs, values or desires; 2, the response to a marketing activity is significantly different from other subgroups; 3, these subgroups have a role in the realization of the objectives of the marketing plan for the enterprise.

Measurement of segmentation standards: 1, measurable: the crowd can be identified; 2, accessible: marketing feasibility (including marketing content and channels); 3, profitable: a certain scale; 4, differentiation; 5, stability

Fourth, the variables used for segmentation

(a) Schiffman to the current selection of Western scholars of the segmentation standards and segmentation Variables are categorized into eight types: geographic segmentation, demographic segmentation, psychological segmentation, socio-cultural segmentation, usage segmentation, usage contextual segmentation, interest segmentation and hybrid segmentation.

ii) But in fact there are three main types of segmentation variables that are generally used in market research:

1. Demographic variables: geographic region (geographic location, overall strength of the city (city line), customs and habits, etc.), age, gender, family size and composition, education, occupation, type of housing, etc.;

2. Behavioral variables: brand role, purchase considerations, usage context, frequency of use, etc.

3. Lifestyle and consumer psychology (also including values, needs and attitudes, etc.; the use of this variable is still the mainstream of market segmentation)

(c) In addition, Wind proposes a model for the selection of segmentation criteria that is oriented to management tasks:

1. Understanding the market as a whole: product purchase and use, demand, brand loyalty, brand switching patterns, etc.

2. Positioning studies: product use, product preferences, benefit seeking, etc.

3. New product introduction: reactions to new product concepts, benefit seeking;

4. Pricing decisions: price sensitivity, price reduction preferences, price sensitivity of different use purchase methods, etc.

p> 5. Advertising decisions: benefits of demand, media use, psychological description/life patterns;

6. Distribution decisions: store loyalty, benefits sought by store choice.

When I first started in the industry, I was told that subjective ratings such as satisfaction and loyalty were not suitable for market segmentation. Now it seems that it is not impossible, but it is necessary to decide which factors to use for market segmentation depending on the industry, the objectives of the study, and the statistical methods that may be used later.

Additionally, I have encountered clients and colleagues who are enamored of models and believe that segmentation without statistical models is crude and unrepresentative. In fact, market segmentation does continue to introduce new statistical methods as statistics evolve. And these methods are nothing more than tools. If some ideas and hypotheses are not put forward in the early stage through characterization, it is difficult to obtain effective market segmentation with even the best statistical tools. Therefore, if the qualitative stage has generated a clear insight that helps the project purpose, it is sufficient to set up relevant questions in the questionnaire to verify it, and it may not be necessary to use a large number of lifestyle/values questions or to do complex statistical analysis models.

V. Market Segmentation Methods

There are many market segmentation methods, such as segmentation factor division, factor clustering, typical correlation clustering, latent class clustering (latent class model), artificial neural networks, and conjoint analysis. Here is a brief record of several segmentation methods:

1. Segmentation factor division method:

Marketers directly choose segmentation factors to artificially divide the overall market into segments. The most commonly used segmentation factors include demographic and behavioral factors. Advantages are simplicity, strong segmentation, easy to identify and describe; disadvantages of using demographic factors are that they do not take into account intrinsic motivation, weak ability to predict behavior, and are not as simple as one might think (many demographic variables are in fact correlated); another advantage of using behavioral factors is that they correlate with consumer behavior, choices, and brand use, which allows for the identification of new market opportunities, but looking at the behavior without looking at the why. Lack of diagnostic value

2, factor clustering:

The use of segmentation factors are mainly for the need, attitude, lifestyle, consumer values and other aspects of the rating title. Problems to be noted are: 1, whether the attitude statements in the questionnaire truly reflect the needs of consumers; 2, because the process of factor clustering does not take into account the demographic characteristics of consumers and consumer behavior, and thus the classification results are often the case that there is a lack of obvious differences in the behavior and background between the segments and the poor identifiability, it is more difficult to provide actionable advice; 3, the problem of internal logic: in fact, it is assumed that the attitudes of different of consumers and their purchasing behavior is different, however, this is not the case.

3. Typical correlation clustering

Because of the problems with factorial clustering mentioned above, it may be necessary to introduce more variables into the model, such as purchase motivation, brand preference, and the influence of advertising and promotion. And these variables are usually not continuous variables that can be used for factor clustering, so typical correlation clustering is introduced. Typical correlation clustering actually replaces factor analysis with typical correlation. The advantage is that it can take into account product needs, values, demographic characteristics and consumer behavior at the same time, so that the final segmentation of the population is intrinsic, logical, and qualitatively interpretable; at the same time, the choice of variables is more flexible, and both qualitative and quantitative variables are acceptable.

4. Description of several clustering methods

1) Hierarchical clustering: can deal with both categorical variables and continuous variables, but cannot deal with both categories at the same time, and does not need to specify the number of categories. There is a nested, or hierarchical, relationship between the clustering results.

2) K-Means Cluster, also known as fast clustering. For continuous variables, can also deal with ordered categorical variables, the operation is very fast, but you need to specify the number of categories. k-means clustering method will not automatically standardize the data, you need to manually standardize the analysis of their own first.

3) Two-Step Cluster: It can deal with both categorical variables and continuous variables, and can automatically identify the best number of categories, and the results are more stable. If only continuous variables are clustered, the distance between records can be described using the Euclidean distance, or log-likelihood (Log-likelihood), if the former, the method and the traditional clustering method is not very different; but if there are also discrete variables clustered, then only log-likelihood can be used to express the differences between records. The difference between the records is expressed using the log-likelihood value. When the clustering indicator is an ordered category variable, the classification result of Two-Step Cluster is not as clear as that of K-means cluster, because the K-means algorithm assumes that the clustering indicator variable is a continuous variable.

4) a few notes on cluster analysis: A, get the factor to be clear whether it is based on the amount of clustering or pattern-based clustering; B, the number of clusters, try between 3-7, spss, if you want to ensure that the samples are sorted beforehand; C, with F-test of the various categories of the clustering variables in the existence of a significant difference; D, test the results of the clustering of different categories: for example, the results of the 3 categories and the 4 categories for the interaction analysis to see where the changes are and what variables are mainly affected; E. For potentially stable clusters, test the F-statistic of each variable (original questionnaire) in each class; F. Write the finalized clustering results into the original dataset, naming it; G. Discriminant analysis, visualize the discriminant classes and clustered variables, and draw discriminant plots to further identify the characteristics of the classes; H. Use correspondence analysis and multivariate Correspondence Analysis and Multiple Correspondence Analysis to identify the attributes of the classes and key class (segmentation) expression variables, such as gender, age, occupation, income, etc.; I. CHAID Classification Decision Tree for automated detection to further identify the characteristics of the classes.

References:

Market Segmentation and Positioning: James H. Myers

Market Research Practices and Methods: Zheng Zongcheng, Chen Jin, and Zhang Wenshuang

An Overview of Market Segmentation Research-Reviews and Perspectives: Luo Jining

Mr. Shen Hao's Blog