Summary: Marketing Research, an applied approach by Malhotra

2. The marketing research problem and approach.

Problem definition: broad statement of the general problem and identification of the specific components of the marketing research problem.

Together with developing an approach, the process of defining the problem contains:
-  Tasks involved:
* discussions with decision maker(s)

Problem audit: comprehensive examination of a marketing problem to understand its origin and nature.
This is important because the decision maker(s)  usually has only a vague idea of what the problem is.

To be productive, the interaction between the decision maker(s) and the researcher should be defined by the 7 c’s:
> Communication: free exchange of ideas.
> Cooperation: it’s a team project.
> Confidence: mutual trust.
> Candor: no hidden agendas and an open attitude.
> Closeness: a warm and closeness relationship.
> Continuity: they must interact continually.
> Creativity: the interaction should be creative instead of formulaic.

* Interviews with experts (to help define the marketing research problem, not to develop a conclusive solution)

Experience survey/Key-informant technique: interviews with people who are very knowledgeable about the general topic being researched.
Lead-user survey: interviews with lead users of the technology.

Expert information is often obtained by unstructured personal interviews, without a formal questionnaire. 

Difficulties by obtaining expert information:
1. Some individuals may not really be experts.
2. It can be difficult to locate and obtain help from experts who are outside the organization.

* secondary data analysis
Secondary data: data collected for another purpose then the problem now.
Primary data: data developed by the researcher specifically to handle the research problem.

(* qualitative research)
Qualitative research: unstructured, exploratory research method based on small samples to get insight and understanding of the problem setting.
Pilot surveys: surveys less structured than large-scale surveys, they generally contain more open-ended questions and have a much smaller sample size.
Case studies: involve an intensive examination of a few selected cases of the symptom of interest. Cases could be customers, stores, etc.
 

-  Environmental context of the problem.
Environmental context of the problem: consists the factors that have an impact on the definition of the marketing research problem, including:

> Past information and Forecasts. This in trends with respect to sales, market share, profitability, technology, population, demographics, and lifestyle.

> Resources and Constraints

> Objectives of the decision maker
Objectives: goals of the organization and of the decision maker must be considered to carry successful marketing research.

> Buyer behavior
Buyer behavior: knowledge that tries to understand and predict customers’ reactions based on an individual’s specific characteristics.

Buyer behavior factors:
- number and geographical location of the buyers and nonbuyers.
- demographic and psychological characteristics.
- product consumption habits and the consumption of related product 
  categories.
- media consumption behavior and response to promotions.
- price sensitivity.
- retail outlets patronized.
- buyer preferences.

> Legal environment
Legal environment: regulatory policies and norms within which organizations must operate.

> Economic environment
Economic environment: consist of income, prices, savings, credit, and general economic conditions.

> Marketing and Technologically skills of the firm.
 

-  Step 1: Problem definition -> management decision problem -> marketing research problem.
Management decision problem: problem confronting the decision maker. It asks what the decision maker needs to do.
Marketing research problem: problem that entails determining what information is needed and how it can be obtained in the most feasible way.

Conceptual map: way to link the broad statement of the marketing research problem to the management decision problem.
It contains these 3 components:
1. Management want to (take an action);
2. Therefore, we should study (topic);
3. So that we can explain (question).

The definition of Marketing Research Problem should (1)allow the researched to obtain all the information needed to address the management decision problem, (2)guide the researched in proceeding with the project.

Broad statement: first statement of the marketing research problem that provides an appropriate perspective on the problem.

Specific components: second part of the marketing research problem which focus on the key aspects of the problem and provide clear guidelines on how to proceed further.
 

-  Step 2: Approach to the Problem:
* Objective/Theoretical Foundations
Theory: conceptual scheme based on foundational statements that are assumed to be true.
Objective evidence: neutral evidence that is supported by empirical findings.

* Analytical Model: Verbal, Graphical, Mathematical
Analytical model: explicit specification of a set of variables and their interrelationships designed to represent some real system or process in whole or part.
Verbal models: analytical models that bring written representation of the relationships between variables.
Graphical models: analytical models that bring a visual picture of the relationships between variables.
Mathematical models: analytical models that explicitly describe the relationships between variables, often in equation form.

* Research questions
Research questions (RQs): refined statements of the specific components of the problem.

* Hypotheses
Hypothesis: unproven statement/proposition about a factor/phenomenon that is of interest of the researcher.

* Specification of Information needed
 

-  Step 3: Research design.

3. Step 3, Research design.

Research design: framework for conducting the marketing research project. It specifies the details of the procedures necessary for obtaining the information needed to structure and/or solve marketing research problems.

A research design involves mostly the following components/tasks:
 

1.  Definition of the information needed. (Chapter A)

2.  Design of the explanatory, descriptive, and/or causal phases of the research. (Chapters B, C, D)

3.  Specification of the measurement and scaling procedures (Chapters E, F)

4.  Construction and pretesting of a questionnaire (interviewing form) or an appropriate form for data collection. (Chapter is not in this summary)

5.  Specification of the sampling process and sample size. (Chapters G, H)

6.  Development of a plan of data analysis. (Chapter is not in this summary)

Classification of the different types of research design:

Exploratory research can be used for the following purposes:
 

-  Formulating a problem or defining a problem more precisely.

-  Indentifying alternative courses of action.

-  Developing hypotheses.

-  Isolating key variables and relationships for further examination.

-  Gaining insights for developing an approach to the problem.

-  Identifying priorities for further research.

There are two types of conclusive research:
 

1.  Descriptive research: major objective is to describe something, usually market characteristics or functions. It’s for the following reasons:
- To describe characteristics of relevant groups.
- To estimate the % of units in a specified population showing a certain
   behavior.
- To determine the perceptions of product characteristics. For example: how
   see households various department stores?
- To determine the degree to which marketing variables are associated. For
   example: to what extent is shopping at department store related to eating out?
- To make specific predictions.

Major methods for descriptive research:

* Secondary data analyzed in a quantitative as counteract to a qualitative
   manner.
* Surveys. (Chapter D)
* Panels (Chapter D)
* Observational and other data (Chapter D)

Sort of descriptive design:
> Cross-sectional design: collection of information from any given sample of population elements only once. They are single or multi:
 

Single cross-sectional design: cross-sectional design in which one sample of respondents is drawn from the target population and information is obtained from this sample once. (also called: sample survey research designs)

Multiple cross-sectional design: there are two or more samples of respondents, and information from each sample is obtained only once. One type of this design:
 

Cohort analysis: consists of a series of surveys conducted at appropriate time intervals. The cohort refers to the group of respondents who experience the same event within the same time interval.

> Longitudinal design: involves a fixed sample of population elements that is measured repeatedly. The sample remains the same over time (is not the case at a cross-sectional design), so providing a series of pictures that, when viewing together, represent a graphic illustration of the situation and the changes that are taking place over time.
 

Panel: sample of respondents who have agreed to provide information at specified intervals over an stretched period.

Advantages and disadvantages of longitudinal an cross-sectional designs:

The non-representative sample of longitudinal design can be because of:
- Refusal to cooperate: some individual or household refuse to participate.
- Mortality/death rate: some people who should participate can move away or lose interest.
- Payment: certain types of people who are attracted.

The response bias can become because of:
- boredom
- fatigue
- incomplete questionnaire entries

2.  Causal research: major objective is to obtain evidence regarding cause-and-effect (causal) relationships. It’s for the following purposed:
- understand which variables are the cause (independent variables) and which are the effect (dependent variables).
- determine the nature of the relationship between the causal variables and the effect to be predicted.

Comparison of basic research designs:

Potential sources of error.

Total error: variation between the true mean value in the population of the variable of interest and the observed mean value obtained in the marketing research project.
 

Random sampling error: error due to the particular sample selected being an imperfect representation of the population of interest. May be defined as the variation between the true mean value for the sample and the true mean value of the population.

Non-sampling error: errors that can be attributed to sources other than sampling, and they can be random or non-random. For example:
 

Non-response error: occurs when some of the respondents included in the sample don’t respond. It may be defined as the variation between the true mean value of the variable in the original sample and the true mean value in the net sample.
 

-  Response error: occurs from respondents who do respond, but give inaccurate answers or their answers are misrecorded/misanalyzed. It may be defined as the variation between the true mean value of the variable in the net sample and the observed mean value obtained in the marketing research project.

Response errors can be made by researchers, interviewers or respondents. Errors made by researches include:
> Surrogate information error: variation between the information needed for the marketing research problem and the information searched by the researcher. (For example: instead of consumers choice, obtains information about consumers preferences)

> Measurement error: variation between the information searched and the information generated by the measurement process employed by the researcher. (For example: while striving for measure of consumer preferences, the researcher made a scale that measures perceptions)

> Population definition error: variation between the actual population relevant to the problem and the population as defined by the researcher.

Budgeting and scheduling: management tools required to help ensure that the marketing research project is completed within the available resources.

A useful approach for this:

Critical Path Method (CPM): management technique of dividing a research project into component activities, determining the sequence of these components and the time each activity will need.
More progressive versions of the CPM:

Program evaluation and review technique (PERT): CPM that accounts for the uncertainty in project completion times.

Graphical evaluation and review technique (GERT): CPM that accounts for both the completion probabilities and the activity costs.

Marketing research proposal: official layout of the planned marketing research activity for management. It contains the essence of the project and serves as a contract between the researcher and management. It describes the research problem, the approach, the research design, data collection methods, data analysis methods, and reporting methods. It gives a cost estimate and a time schedule for completing the project.

Most proposals covers all steps of the marketing research process and contain:

1.  Executive summary: summary of the major points from each of the other sections, presents an overview of the entire proposal.

2.  Background.

3.  Problem Definition/Objectives of the Research.

4.  Approach to the Problem.

5.  Research Design: whether exploratory, descriptive or causal. Information must provide the following components: 1.Kind of information to be obtained. 2.Method of administering the questionnaire (mail, telephone, personal/electronic interviews) 3.Scaling techniques 4.Nature of the questionnaire (type of questions, length) 5.Sampling plan and sample size.

6.  Fieldwork/Data Collection.

7.  Data analysis: simple cross-tabulations, univariate analysis, multivariate analysis and how the results will be interpreted.

8.  Reporting.

9.  Cost and Time. (CPM or PERT may be included)

10.  Appendices: statistical or other information that is only of interest to a few people.

Advantages of making a research proposal:

-  Ensures that the researcher and management agree about the nature of the project.

-  Helps sell the project to management.

5. Qualitative Research in the Exploratory Research Design.

Qualitative research: unstructured exploratory research methodology based on small samples that provides insights and understanding of the problem.

Quantitative research: strives to quantify the data and, typically, applies some form of statistical analysis.

Why use qualitative research? Because it’s not always possible/desirable to use fully structured or formal methods to obtain information from respondents. People may be unwilly or unable to answer certain questions.

Qualitative research classified in:
 

-  Direct approach: the purposes of the project are disclosed to the respondent or are obvious, given the nature of the interview.
-> Focus Groups, Depth Interviews.
 

-  Indirect approach: the purposes of the project are secret to the respondents.
-> Projective Techniques -> Association, Completion, Construction, Expressive Techniques.

Focus group: interview leaded by a trained moderator among a small group of respondents in an unstructured and natural manner.

Main purpose: insights by listening to a group of people from the appropriate target market talk about issues of interest to the researcher.

Characteristics of Focus Groups:

Group size:     8-12

Group composition:    Homogeneous; respondents prescreened

Physical setting:     Relaxed informal atmosphere

Time duration:    1-3 hours.

Recording:   Use of audiocassettes and videotapes

Moderator:   Observational, interpersonal, and communication skills of the

moderator.

Procedure for planning and conducting Focus Groups:
 

-    Determine the objectives of the marketing research project and define the problem.

-    Specify the objectives of qualitative research.

-    State the objectives/questions to be answered by focus groups.

-    Write a screening questionnaire.

-    Develop a moderator’s outline.

-    Conduct the focus group interviews.

-    Review tapes and analyze the data.

-    Summarize the findings and plan follow-up research or action.

Variations in Focus Groups:
 

-  Two-way focus group: allows one target group to listen to and learn from a related group.

-  Dual-moderator group: focus group interview conducted by two moderators. One for the smooth flow of the session, the other for  that the specific issues are discussed.

-  Dueling-moderator group: also two moderators, but they take intended opposite positions on the issues to be discussed.

-  Respondent-moderator group: the moderator asks selected participants to play the role of moderator temporarily to improve group dynamics.

-  Client-participant groups: client personnel are identified and made part of the discussion group. Their primary role is to offer clarifications for a more effective group process.

-  Mini-groups: a moderator and only 4-5 respondents. Used if required more extensive probing.

-  Telesessions: focus group technique using a telecommunications network.

-  Electronic group interviewing (EGI): keypads and other electronic mottos are used to measure group opinion.

Advantages of Focus Groups:
 

-  Synergism: a group of people together will produce a wider range of information, insights and ideas than individual responses privately.

-  Snowballing(bandwagon): one person’s comment triggers a chain reaction from the other participants.

-  Stimulation: often after a brief introductory period, the respondent want to express their ideas and expose their feelings as the general level of excitement over the topic increases in the group.

-  Security: the participants’ feelings are similar to those of the group members, so they feel comfortable and are therefore willing to express their ideas and feelings.

-  Spontaneity: participants are not required to answer specific questions, so their responses can be spontaneous.

-  Serendipity: ideas are more likely to be out of the blue in a group.

-  Specialization: because a number of participants are involved simultaneously, use of a highly trained, but expensive, interviewer is justified.

-  Scientific scrutiny: it allows close scrutiny of the data-collection process: observers can attend the session and it can be recorded for later analysis.

-  Structure: allows for flexibility in the topics and the depth.

-  Speed: data collection and analysis goes relatively quickly because a number of individuals are being interviewed at the same time.

Disadvantages of Focus Groups:
 

-  Misuse: misuse of abuse of it by considering the results as conclusive instead of exploratory.

-  Misjudge: results of focus groups can be more easily misjudged than the results of other data-collection techniques. They are more sensitive to client and researcher biases.

-  Moderation: focus groups are difficult to control. Moderators with all the desirable skills are seldom. The quality of the results depends heavily on the skills of the moderator.

-  Messy: the unstructured nature of the responses makes coding, analysis, and interpretation difficult.

-  Misrepresentation: the results are not representative of the general population and are not projectable. They are not the basis for decision making.

Focus groups can be used in almost any situation that needed some opening understanding and insights.
 

Advantages Online Focus Groups:
 

-  All people over the world can participate, and the client can observe the group from the ease of the home or office. So less geographical and time constraints.

-  The unique opportunity to contact group participants again at a later date

-  Internet enables to reach segments that are often hard to inverview: doctors, lawyers, professionals.

-  Ability for side conversations with individual respondents, probing deeper into interesting areas.

-  People are less banded in their responses and are more likely to fully express their thoughts.

-  Lower costs. (not travel, videotaping etc)

Disadvantages Online Focus Groups:
 

-  Only people who are familiar with a computer can participate.

-  Verifying that a respondent is really a member of a target group is difficult.

-  Body language, facial expressions and tone of voice can’t be obtained, electronic emotions don’t capture totally emotion as a videotape does.

-  Lack of general control over the respondent’s environment and their potential exposure to distracting external stimuli. The researcher don’t know what a participant does while participating.

-  Only audio and visual stimuli  can be tested.

-  Products cannot be touched (clothes) or smelled (perfumes).

Depth interview: unstructured, direct personal interview in which a single respondent is probed by a highly skilled interviewer to uncover underlying motivations, beliefs, attitudes, and feelings on about a topic. (qualitative research)

Techniques of depth-interviewing:
 

1.  Laddering: a line of questioning proceeds from product characteristics to user characteristics. It is a way to probe into customers’ deep underlying phychological and emotional reasons. The line of questioning generally proceeds from product characteristics to user characteristics. Following initial responses with ‘why’ questions leads to much more useful information.
 

2.  Hidden issue questioning: attempts to locate personal sore spots related to deeply felt personal concerns. It is not focused on socially shared values and generally lifestyles.
 

3.  Symbolic analysis: the symbolic meaning of objects is analyzed by comparing them with their opposites.

Advantages of depth interviews:
 

-  Uncover greater depth and insights than focus groups.

-  It attributes the responses directly to the respondent (in focus groups it is often difficult to determine which respondent made a particular response)

-  Free exchange of information because there is no social pressure to merge to group response.

Disadvantages of depth interviews:
 

-  See the disadvantages of focus groups (and often to a greater extent)

-  Expensive and difficult to find.

Variation of depth interviews:
 

-  Grounded theory: inductive and more structured approach in which each subsequent depth interview is adjusted based on the cumulative findings from previous depth interviews with the purpose of developing general concepts/theories.

-  Protocol interview: a respondent is placed in a decision-making situation and asked to put into words the process and the activities that he/she would undertake to make the decision.

Projective techniques: unstructured and indirect form of questioning that stimulates the respondents to project their underlying motivations, beliefs, attitudes, or feelings regarding the issue of concern.

Types of projective technique:
 

-  Association techniques: the respondent is presented with a stimulus and asked to respond with the first thing that comes to mind. Like:

>Word association: respondents are represented with a list of words, and after each word they are asked to give the first word that comes to mind. (the words are one at a time presented)

These techniques are used to reveal from the correspondent their inner feelings about the topic of interest.

Responses are analyzed by calculating:

1.Frequency of any word that is given as a response

2.Amount of time before a response is given

3.Number of respondents who have not responded at all.

-  Completion technique: the respondent must complete and incomplete stimulus situation. Like:

>Sentence completion: respondents are presented with a number of incomplete sentences and asked to complete them.

>Story completion: respondents are provided with part of a story and required to give the conclusion in their own words.
 

-  Construction technique: respondent is required to construct a response in the form of a story, dialogue, or description. The two main construction techniques are:

1. Picture response technique: respondent is shown a picture and asked to tell a story describing it.

2. Cartoon tests: cartoon characters are shown in a specific situation related to the problem. The respondents are asked to indicate the dialogue that one cartoon character might make in response to the comments of the other character.

-  Expressive techniques: respondent is presented with a verbal or visual situation and asked to relate the feelings and attitudes of other people to the situation.
The two main expressive techniques:

1. Role playing: respondents are asked to assume the behavior of someone else.The researcher assumes the respondent will project their own feelings into the role.

2. Third-person technique: the respondent is presented with a verbal or visual situation and asked to relate the beliefs and attitudes of a third person to the situation.

Advantages of projective techniques:
 

-  Cause responses that would be unwilling or unable to give if they knew the purpose of the study. In direct questions (focus groups and depth interviews) the respondent may misunderstand, misinterpret or mislead the researcher.

-  Helpful for underlying motivations, beliefs, and attitudes.

Disadvantages of projective techniques:
 

-  Unstructured direct techniques require generally personal interviews with highly trained interviewers.

-  Risk of interpretation bias. Except for the word association, all techniques are open ended, which makes the analysis and interpretation difficult and subjective.

-  Some (like role playing) requires the respondents to adopt in unusual behaviors, which may not be representative of the population of interest.

Projective techniques are less used than unstructured direct methods. Except for word association which is commonly used to test brand names and attitudes about particular products, brands, advertisements.

Qualitative data: words are the units of analysis. The goal of qualitative research is to solve, examine, and interpret meaningful patterns/themes that emerge out of data. The three general steps followed in analyzing qualitative data:
 

1.  Data reduction: choose which aspects of the data are accented, minimized for the project.

2.  Data display: develop a visual interpretation of the data with the use of tools suchs as a diagram, chart or matrix.

3.  Conclusion drawing and verification: think over the meaning of the analyzed data and estimate it for the research question.

Software packages are available for use to assist in the analysis of qualitative data. There are six main types:

-  Word processors

-  Word retrievers

-  Text-base managers

-  Code-and-retrieve programs

-  Code-based theory builders

-  Conceptual network builders

The specific things that various programs can do:

1.  Coding.

2.  Memoing/annotation: it allows you to make side notes that correspond to sections of your data.

3.  Data linking.

4.  Search and Retrieval: it allows you to search for specific words.

5.  Conceptual/Theory Development.

6.  Data Display: shows results onscreen or even with split screens.

7.  Graphics Editing.

Qualitative research is crucial in international marketing research, because researchers are often not familiar with the foreign product market to be examined. It may reveal the differences between the foreign and domestic market.

6. Survey and Observation in the Descriptive Research Design

Survey method: structures questionnaire given to a sample of a population and designed to get specific information from respondents.

Structured data collection: use of a formal questionnaire with questions in a prearranged order.

Whether research is classified as direct or indirect is based on whether the true purpose is known to the respondents.

The most popular data collection method is the structured-direct survey, which involves administering a questionnaire. Most questions here are fixed-alternative questions: require respondents to choose from a set of predetermined answers. For example:

     Disagree       Agree

Shopping in department stores sucks   1   2   3   4   5

Advantages of the survey method:
 

-  Questionnaire is simple to administer.

-  The obtained data are reliable because the responses are limited to the alternatives given.

-  Reduces the variability in the results.

-  Coding, analysis and interpretation are relatively simple.

Disadvantages of the survey method:
 

-  Respondents may be unable/unwilling to provide the desired information, not giving accurate answers.

-  Respondents may be unwilling to respond by sensitive or personal questions.

-  Loss of validity for certain types of data such as beliefs and feelings.

-  Wording question is properly not easy.

Survey questionnaires may be administered in four major modes:
 

1.  Telephone methods:

* Traditional: phoning a sample of respondents and asking them a series of questions.
* Computer-Assisted (CATI): uses a computerized questionnaire administered to respondents over the telephone.
 

2.  Personal interviews:

*In-Home: respondents are interviewed face-to-face in their homes.
*Mall Intercept: respondents are intercepted while they are shopping in malls/centers and brought to test facilities in the malls. The interviewer then administers a questionnaire.
* Computer-Assisted (CAPI): respondents sits in front of a computer terminal and answers questionnaire on the computer screen by using keyboard/mouse.
 

3.  Mail interviews:

* Mail: questionnaires are mailed to preselected potential respondents.
* Mail Panels: large and nationally representative sample of households who have agreed to periodically participate in mail questionnaires, product tests, and telephone surveys.
 

4.  Electronic interviews:

* E-Mail: for an e-mail survey, first a list of e-mail addresses is obtained. The survey is written in the e-mail message and the e-mails are sent out over the Internet. (use pure text: ASCII) The questionnaires can be received and responded by anyone with an e-mail address.
* Internet: different from e-mail surveys, Internet/Web surveys use hypertext markup language (HTML): the language of the web. They are posted on a Website. Respondents may be recruited on the Internet from potential respondent databases, an Internet Panel, or by traditional methods (mail, telephone). Then they have to go to a particular Web location to complete the survey.

In table 6.2 on page 222: a comparative evaluation of survey methods on the basis of the criteria:
 

-  Task Factors: tasks that have to be performed to collect the data and to the topic of the survey:
* diversity of questions and flexibility.

* use of physical stimuli, such as the product, product proto type, commercials,
   etc.

* sample control: ability of the survey mode to reach the units specified in the
   sample effectively and efficiently.
   Sampling framing: list of the elements of the target population units with
   their telephonenumbers.
   Random digit dialing (RDD): technique used to win over the bias of
   unpublished and recent telephone numbers by selecting all telephone number
   digits at random. (it does not distinguish between telephone numbers that are
   interested and not)
   Random digit directory designs: research design for telephone surveys in
   which a sample of numbers is drawn from the telephone directory and
   modified to allow unpublished numbers a change of being included in the
   sample.

* quantity of data: amount of data collected.

* response rate: percentage of the total attempted interviews that are
   completed.
   Non-response bias: when actual respondents differ from those who refuse to
   participate. (low response rate)

   Multiple request strategies: a small first request followed by a larger request,
   called the critical request: the target behavior. This to increase the response
   rate.

For example: higher amount of monetary incentive, nonmonetary
   premiums and rewards (pens, books), foot-in-the-door techniques)
 

-  Situational Factors:
* control of data collection environment: control of the researcher over the
   environment of the respondent during answering the questionnaire.

* control of field force: interviewers and the supervisors involved in data
   collection.

* potential for interviewer bias. For example in the way he/she selects
   respondents, asks research questions, records answers.

* speed: fast method? (like the Internet)

* costs.
 

-  Respondent Factors: relates to survey respondents and contains:

* perceived anonymity: respondent’s perceptions that their identities will not
   be differentiate by the interviewer or the researcher. Is high in mail surveys
   and so on, because there is no contact with the interviewer while answering.

* social desirability: tendency of the respondents to give answers that may not
   be accurate but that may be desirable from a social standpoint.

* obtaining sensitive information

* low incidence rate: rate of occurrence or the percentage of persons
   preferable to participate in the research: how many contacts need to be
   screened for a given sample size requirement.

* respondent control: when answering the survey and the flexibility to answer it
   in parts at different times.

Some other survey methods:

Observation: recording of behavioral patterns of people, objects, and events in a systematic manner to obtain information about the thing of interest.

Observation methods: (second type of methodology used in descriptive research)
 

-  > Structured observation: the researcher clearly defines the behaviors to be observed and the methods by which they will be measured.

> Unstructured observation: researcher monitors all relevant phenomena without specifying before the details.
 

-  > Disguised observation: respondents don’t know that they are observed.
> Undisguised observation: respondents know they are observed.
 

-  > Natural observation: observing behaviors as it takes place in the environment.

> Unnatural observation: observing behaviors in an artificial environment.

Observation methods classified by Mode of Administering:
 

-  Personal observation: human observers observe the phenomenon as it occurs.
 

-  Mechanical observation: mechanical equipment instead of human researchers observe the phenomenon.

Psychogalvanometer: instrument that measures a respondent’s galvanic skin response.
Galvanic skin response (GSR): changes in the electrical resistance of the skin that relate to a respondent’s affective state.

Voice pitch analysis: measuremen of emotional reaction through changes in the respondent’s voice.

Response latency: amount of time it takes to respond. (respond time)
 

-  Audit: researcher collects data by examining physical records or performing inventory analysis. It has two distinguish characteristics:
1. Data are collected by the researcher himself.
2. Data are based upon counts, often of physical objects.

Pantry audit: the researcher inventories the brands, quantities, and package size of products in a consumer’s house.
 

-  Content analysis: objective, systematic, and quantitative description of the manifest content of a communication. (is an appropriate method when the phenomenon to be observed is communication instead of behavior or physical objects) It includes observation and also analysis.
 

-  Trace analysis: data collection is based on physical traces, or evidence, of past behavior.

Comparative evaluation of Observation Methods:
 

Only 1% of the marketing research projects are solely based on observational methods to obtain primary data. So observational methods have some major disadvantages in comparison to survey methods.

Relative advantages of observations:
 

-  They permit measurement of actual behavior instead of reports of intended or preferred behavior.

-  No reporting bias and lower bias by the interviewer and the interviewing process.

-  Certain types of data can only be collected by observation: if the respondent is unaware or unable to communicate. For example information about preferences for babies’ toys.

-  May be cheaper and faster than survey methods, when the observed phenomenon occurs frequently or is of short duration.

Relative disadvantages of observations:
 

-  The reasons for the behavior are not determined.

-  Selective perception (bias in the researcher’s perception) can bias the data.

-  It is time-consuming and expensive, and difficult to observe certain forms of behavior like personal activities.

-  Sometimes it may be unethical, as in monitoring the behavior of people without their knowledge about it.

Ethnographic research is the study of human behavior in its natural context and involves observation of behavior and setting along with depth interviews. So the questioning and observation methods are combined to understand the behavior.

Mystery shopping: trained observers pose as consumers and shop at company (or competitor) owned stores to collect data about customer and employee interaction and other marketing variables (prices, displays, answers, etc).

8. Fundamentals and Comparative Scaling (Measurement and Scaling).

Measurement: assignment of numbers or other symbols to characteristics of objects according to certain specified rules.

Scaling: generation of a continuum upon which measured objects are located.

Scale characteristics:

-  Description: the unique labels or descriptors used to name each value of the scale. For example: 1 = female, 2 = male.
 

-  Order: the relative sizes or positions of the descriptors. It is denoted by descriptors like greater than, less than, and equal to.
 

-  Distance: absolute differences between the scale descriptors are known and may be expressed in units. For example: number of persons living in your household.
 

-  Origin: the scale has a unique or fixed beginning or true zero point. For example: annual income starts at zero $.
A scale with origin has also distance (and order and description).

Primary scales of measurement: (Figure 8.1, page 284)

1.  Nominal scale: the numbers serve only as labels for identifying and classifying objects. If used for identification, there is a strict-to-one correspondence between the numbers and the objects. For example: number assigned to runners.
 

2.  Ordinal scale: ranking scale in which numbers are assigned to objects to indicate the relative extent to which some characteristics is possessed. So you can determine whether an objective has more or less of a characteristic than some other object. For example: rank orders of winners.
 

3.  Interval scale: numbers are used to rate objects such that numerically equal distances on the scale represent equal distances in the characteristic being measured. For example: performance rating on a scale 0-10: 8.2 – 9.1 – 9.6.
(difference 1-2 is the same as 3-4)
 

4.  Ratio scale: the highest scale, to identify or classify objects, rank-order the objects, and compare intervals or differences. It is also meaningful to compute ratios of scale values. It possesses all the properties of the nominal, ordinal and interval scales, and an absolute zero point. For example: time to finish in sec. 15.2 – 14.1 – 13.4.
There is not only 1-2 is the same as 13-14, but also 14 is 7 times as large as 2 in absolute sense.

Scaling techniques:

-  Comparative (non-metric) scales: there is direct comparison of stimulus objects with one another:

> Paired Comparison scaling: respondent is presented with two objects at a time and asked to select one object in the pair according to some criterion. The obtained data are in nature ordinal

Transitivity of preference: assumption made to trade paired comparison data to rank order data. If brand A is preferred to B, and B to C, then A is preferred to C.

> Rank Order scaling: respondents are presented with several objects simultaneously and asked to order or rank them according to some criteria.

> Constant Sum scaling: respondents are required to allocate a constant sum of units such as points, dollars, stickers, etc. among a set of stimulus objects according to some criteria. For example: allocate 100 points to soap characteristics.

> Q-Sort and Other Procedures. Q-Sort scaling: uses a rank order procedure to sort objects based on similarity according to some criteria. For example: 100 attitude statements on individual cards and place them into 11 piles, ranging from ‘most highly agreed with’ to ‘least highly agreed with’. The number of objects to be sorted should not be less than 60 nor more than 140.

Magnitude estimation: numbers are assigned to objects such that ratios between the assigned numbers reflect ratios on the specified criteria. For example: indicate whether agree or disagree and then assign a number between 0-100 to each statement to indicate the intensity of their agreement or disagreement.

Guttman scaling (scalogram analysis): determines whether a set of objects can be ordered into an internally consistent, unidimensionale scale.

Advantages of comparative scales:
- small differences between stimulus objects can be detected, and respondents
- approach the rating task from the same known reference points.
- involve fewer theoretical assumptions
- tend to reduce halo or carryover effects from one judgment to another.

Disadvantages of comparative scales:
- the ordinal nature of the data
- inability to generalize beyond the stimulus objects scaled.
 

-  Non-comparative (monadic/metric) scales: each stimulus is scaled independently of the other objects in the stimulus set.
The resulting data are mostly assumed to be interval or ratio scaled. For example, rate preference for cola 1-6, 1 = not at all preferred, 2 = etc.)

> Continuous Rating Scales.

> Itemized Rating Scales:
   * Likert
   * Semantic Differential
   * Stapel

9. Non-comparative scaling Techniques (Measurement and Scaling).

Non-comparative scale: each stimulus is scaled independently of the other objects in the stimulus set.

> Continuous (graphic) rating scale: respondents rate the objects by placing a mark at the appropriate position on a line that runs from one extreme of the criterion variable to the other.
For example: The worst ----------------|-------------------------------------------------------The best

> Itemized Rating Scale: measurement scale having numbers and/or brief descriptions associated with each category. The categories are ordered in terms of scale position:

   * Likert scale: has 5 response categories ranging from ‘strongly disagree’ to ‘strongly agree’, which requires the respondent to indicate a degree of agreement or disagreement with each of the series of statements related to the stimulus objects.

   * Semantic Differential: 7-point rating scale with endpoints associated with bipolar (two-sided) labels that have semantic meaning. For example:
Cold _._._._._._._ Warm

   * Stapel scale: consist of a single adjective in the middle of an even-numbered range of values, from -5 to +5, without a neutral (zero) point.
For example: +5 +4 +3 +2 +1 High Quality -1 -2 -3 -4 -5

The basic non-comparative scales:

Decisions in constructing an non-comparative Itemized Rating Scale:

1.  Number of scale categories.
-> although there is no single, optimal number, traditional guidelines suggest there should be between 5-9 categories.
 

2.  Balances vs. unbalanced scale.
Balanced scale: the number of favorable and unfavorable categories are equal.

-> In general, the scale should be balanced to obtain objective data.
 

3.  Odd or even number of categories.
Odd number of categories: the middle point is often neutral.
 

4.  Forced vs. non-forced choice.
Forced rating scale: forces the respondents to express an opinion because ‘no opinion’ or ‘don’t know’ is not given.
 

5.  Nature and degree of the verbal description: can be verbal, numerical or even pictorial description and the researcher has to label or every scale category, some scale categories or only extreme scale categories.

-> the category descriptions should be located as close to the response categories as possible.
 

6.  Physical form of the scale: scales presented horizontally or vertically, categories expressed in boxes, lines or units on a continuum, etc.

Multi-item scales: consists of multiple items, where an item is a single question or statement.

The development of multi-item rating scales requires technical expertise.
The researcher begins by developing the construct of interest. Construct: specific type of concept that exists at a higher level of abstraction than do everyday concepts.

A multi-item scale must be evaluated for accuracy and applicability, so it involves assessment of reliability, validity, and generalizability: (Fig.9.5 page 317)

A measurement is not the true value of the characteristic, but an observation of it. Measurement error: observed score is different from the true score.
True score model: framework to understand the accuracy of measurement:
Xo = Xt + Xs + Xr
where:

Xo = observed score
Xt = true score
Xs = Systematic error: affects the measurement in a constant way and represents
    stable factors that affect the observed score in the same way
    each time the measurement is made.
Xr = Random error: arises from random changes or differences in respondents or
   measurement situations.

Scale Evaluation:
 

-  Reliability: extent to which a scale produces consistent results if repeated. Systematic error don’t have an adverse impact on reliability, since they affect it in a constant way. Reliability is the extent to which measures are free from random error. If Xr = 0, the measure is perfectly reliable.

* Test/Retest reliability: respondents are administered identical sets of scale items at two different times under equivalent conditions.

* Alternative Forms reliability: requires two equivalent forms of the scale to be constructed and then the same respondents are measured at two different times.

* Internal Consistency reliability: several items are summated in order to form a total score for the scale. Sorts:

   > Split-half reliability: the items on the scale are divided into two halves and
   the resulting half scores are correlated. High correlation means high
   internal consistency. The results depend on how the scale items are split.

   To overcome this problem use:
 
  Coefficient (Cronbach’s) alpha: average of all possible split-half
   coefficients resulting from different splittings of the scale items. It varies 0
   and 1, and value of 0,6 or less often indicates unsatisfactory internal
   consistency reliability.
   Coefficient beta: to determine whether the averaging process used in
   calculating coefficient alpha is masking any consistent items.
 

-  Validity: extent to which differences in observed scale scores reflect true differences among objects on the characteristics being measured, instead of systematic or random errors. Perfect validity: there is no measurement error: Xo = Xt, Xr = 0, Xs = 0

* Content (face) validity: consists of a subjective but systematic evaluation of
    the representativeness of the content of a scale.
   The researcher examines if the scale items adequately cover the entire
    domain of the construct. A content validity alone is not sufficient to measure
    validity.

* Criterion validity: examines if the measurement scale performs as expected
   in relation to other variables selected as meaningful criteria. These criterion
   variables may include demographic and psychographic characteristics,
   attitudinal and behavioral measures, or scores obtained from other scales.
  
    It can take two forms:

   > Concurrent validity: when the data on the scale being evaluated and on the
  criterion variables are collected at the same time.

   > Predictive validity: researcher collects data on the scale at one point in time
  and data on the criterion variables at a future time.

* Construct validity: addresses the question of what construct or characteristic
   the scale is measuring. And made an attempt to answer theoretical questions
   about why the scale works and what deductions can be made concerning the
   theory underlying the scale.

   > Convergent validity: extent to which the scale correlates positively with
  other measures of the same construct.

   > Discriminant validity: extent to which a measure does not correlate with
  other constructs from which it is supposed to differ. It demonstrates a lack of
  correlation among different constructs.

   > Nomological validity: extent to which the scale correlates in theoretically
  predicted ways with measures of different but related constructs.
  A theoretical model is formulated that leads to further deductions, tests, and
  inferences.
 

-  Generalizability: degree to which a study based on a sample applies to a universe of generalizations.

If  a measure is perfectly valid, it is also perfectly reliable.

If a measure is unreliable, it cannot be perfectly valid, so unreliability implies invalidity.

If a measure is perfectly reliable it may or may not be perfectly valid, because systematic error may still be present.

11. Design and Procedures (Sampling)

Population: sum of all the elements, sharing some common set of characteristics, that is the universe for the purpose of the marketing research design.

Census: complete numeration of the elements of a population or study objects.

Sample: subgroup of the elements of the population selected for participation in the study.

Conditions favoring the use of:

5 Steps in the Sampling Design Process:
 

1.  Define the target population.
Target population: collection of elements that possess the information sought by the researcher and about which inferences are to be made.
The target population should be defined in terms of elements, sampling units, extent, and time.
Element: object that possesses the information sought by the researcher.
Sampling unit: basic unit containing the elements of the population to be sampled.
 

2.  Determine the sampling frame.
Sampling frame: representation of the elements of the target population. Consists of a list or set of directions for identifying the target population.
 

3.  Select a sampling technique(s):
- use a Bayesian or traditional sampling approach?
  Bayesian approach: elements are selected sequentially. It incorporates
   explicitly prior information about population parameters as well as the costs
   and probabilities associated with making wrong decisions.
 
   Traditional sampling: the entire sample is selected before data collection
   starts. Is most commonly used. (information about costs and probabilities is
   rare).

  Sampling with replacement: sampling technique in which an element can be
   included in the sample more than once.

  Sampling without replacement: an element cannot be included in the 
   sample more than once.

The most important decision in selecting sampling technique(s) is whether to use probability or non-probability sampling.
 

4.  Determine sample size.
Sample size: number of elements to be included in the research. Depends on:
- importance of the decision
- nature of the research
- number of variables
- nature of the analysis
- sample sizes used in similar studies
- incidence rates
- completion rates
- resource constraints
 

5.  Execute the Sampling Process.
Requires detailed specification of how the sampling design decisions regarding to the population, sampling frame, sampling unit, sampling technique, and sample size are to be implemented.

Sampling techniques:

-    Non-probability sampling: don’t use chance selection procedures, but rely on the personal judgment of the researcher

-> Convenience Sampling: attempts to obtain a sample of convenient
  elements. The selection of sampling units is primarily left to the interviewer.
  Respondents are often selected because they are in the right place at the
  right time.

-> Judgmental Sampling: the population elements are purposely selected
  based on the judgment of the researcher.

-> Quota Sampling: is a two-stage restricted judgmental sampling. First stage:
  developing control categories or quotas of population elements. (sex, age)
  Second stage: selection of sample elements based on convenience or
  judgment.

-> Snowball Sampling: an initial group of respondents is selected randomly.
  Subsequent respondents are selected based on the referrals or information
  provided by the initial respondents. This process may be carried out in
  waves by obtaining referrals from referrals. (to get characters that are rare)
 

-   Probability sampling: each element of the population has a fixed probabilistic chance of being selected for the sample. (Fig.11.4 page 382)

-> Simple Random Sampling (SRS): each element in the population has a known and equal probability of selection. Every element is selected independently of every other element and the sample is drawn by a random procedure from a sampling frame.

-> Systematic Sampling: sample is chosen by selecting a random starting point and then picking every i-th item in succession from the sampling frame.

-> Stratified Sampling: two-step process to partition the population unto
subpopulations, or strata. Elements are selected from each stratum by a random procedure.

    * Proportionate: the sample size of each stratum is proportionate to the
  relative size of that stratum in the total population.
    * Disproportionate: sample size from each stratum is proportionate to the
  relative size of that stratum and to the standard deviation of the distribution
  of the characteristics of interest among all the elements in that stratum.

-> Cluster Sampling: First the target population is divided into mutually exclusive and collectively exhaustive subpopulations: clusters. Second a random sample of clusters is selected based on a probability sampling technique such as simple random sampling. For each selected cluster, either all the elements are included in the sample or a sample of elements is drawn probabilistically.

Cluster sample can be One-Stage, Two-Stage, and multistage.
One-stage: if all elements in each selected cluster are included in the sample.
Two-stage: if a sample of elements is drawn probabilistically from each selected
    cluster.
Multistage: more than two stages.

A common form of cluster sampling is area sampling: clusters consists of geographic areas, such as countries, housing blocks.

Single-stage area sampling: if only one level of sampling takes place in selecting the basic elements. So you have also two-stage and multistage area sampling.
This design is useful when the clusters are equal in size. If not:

Probability proportionate to size sampling (PPS): the clusters are selected with probability proportional to size and the probability of selecting a sampling unit in a selected cluster varies inversely with the size of the cluster. So first, large clusters are more likely to be included than small clusters. Second, the probability of selecting a sampling unit in a selected cluster varies inversely with the size of the clusters.

-> Other Sampling Techniques:
Sequential sampling: the population elements are sampled sequentially, data collection and analysis are done at each stage, and a decision is made as to whether additional population elements should be sampled.

Double sampling: certain population elements are sampled twice.

Differences between stratified and cluster sampling:

Strengths and weaknesses of Basic Sampling Techniques:

Internet Sampling:

-    Online Intercept Sampling:
-> non-random
-> random
 

-    Recruited Online Sampling:
-> Panel:
    * Recruited Panels
    * Opt-in Panels
-> Non-panel:
    * Opt-in List Rentals
 

-    Other Techniques

To maintain reliability and integrity in the internet sample:

-  Password protection.

-  Reminder invitations.

-  Summary of the survey findings. 

12. Final and Initial Sample Size Determination (Sampling)

Parameter: summary description of a fixed characteristic or measure of the target population. It denotes the true value that would be obtained if a census instead of a sample were undertaken.

Statistic: summary description of a characteristic or measure of the samle.

Finite population correction (fpc): correction for overestimation of the variance of a population parameter.

Precision level: desired size of the estimating interval.

Confidence interval: range into which the true population parameter will fall, assuming a given level of confidence.

Confidence level: probability that a confidence interval will include the population parameter.

Sampling distribution: distribution of the values of a sample statistic computed for each possible sample that could be drawn from the target population under a specified sampling plan.

Statistical inference: process of generalizing the sample results to the population results.

The important properties of the sampling distribution of the mean, and the corresponding properties for the proportion, for large samples (30 or more) are:

1.  Normal distribution: basis for classical statistical inference that is bell-shaped and symmetrical in appearance. Its measures of central tendency are all identical.

2.  Mean:    Formule 1 (zie bijlage)   Proportion: P = X/n

3.  Standard deviation is called the standard error of the mean or the proportion.
Standard error: standard deviation of the sampling distribution of the mean or proportion.  The formulas are:  Formule 2 (zie bijlage)
 

4.  The population standard deviation, σ, is not known. In this case:

Formule 3 (zie bijlage)

5.  Z-value: number of standard errors a point is away from the mean.

Formule 4 (zie bijlage)

Confidence interval:

If the sample size is 10% or more of the population size, the standard error formulas will overestimate the standard deviation of the population mean or proportion. So this should be adjusted by a finite population correction factor defined by: (and σ known)

  Formule 5 (zie bijlage)

So, we can see that:   Formule  6 (zie bijlage)

Incidence rate: rate of occurrence of persons eligible to participate in the study expressed as a percentage.

Incidence rate = Q1 * Q2 * Q3 … * Qc

Completion rate: percentage of qualified respondents who complete the interview. It enables researchers to take into account anticipated refusals by people who qualify.

 Initial sample size = Final sample size / (Incidence * Completion rate)

Response Rate = Number of Completed Interviews / Number of Eligible Units in Sample

The two major non-response issues in sampling are:

-  Improving response rates

-  Adjusting for non-response.

Methods for Improving Response Rates:
 

-  Reducing Refusals:
* prior notification
* motivating respondents
* incentives
* questionnaire design and administration
* follow-up: contacting the non-respondents periodically after the initial contact.
* other facilitators, for example personalization: send letters addressed to
   specific individuals.
 

-  Reducing Not-at-Homes
* callbacks

Strategies to adjust for non-response error:

-  Sub-sampling of non-respondents: results often in a high response rate within the sub-sample.
 

-  Replacement: current non-respondents replaced with non-respondents from an earlier, similar survey.
 

-  Substitution: substitutes for non-respondents other elements from the sampling frame that are expected to respond.
 

-  Subjective estimates: evaluating the likely effects of non-response based on experience and available information. For example adults with children are more likely to be at home than a single worker without children.
 

-  Trend analysis: researcher tries to distinguish a trend between early and late respondents. It is projected to non-respondents to estimate their characteristic of interest.
 

-  Simple weighting: attempts to account for non-response by assigning differential weights to the data depending on the response rate.
 

-  Imputation: assigning the characteristic of interest to the non-respondents based on the similarity of the variables available for both non-respondents and respondents. For example, a respondent who did not reported brand usage may be imputed the usage of a respondent with similar demographic characteristics.

15. Frequency Distribution, Cross-Tabulation, and Testing of Hypothesis.

Frequency distribution: mathematical distribution whose goal is to obtain a count of the number of responses associated with different values of one variable and to express these counts in percentage terms.

The most commonly used statistics associated with frequencies are:

-  Measures of Location: statistic that describes a location within a data set. Measures of central tendency describe the center of the distribution. Measures of location:

* Mean: average. Value obtained by summing all elements in a set and dividing by the number of elements.

* Mode: measure of central tendency given as the value that occurs the most in a sample distribution.

* Median: measure of central tendency given as the value above which half of the values fall and below which half of the values fall.
 

-  Measures of variability: indicates the distribution’s dispersion/spread.

* Range: difference between the largest and smallest value of a distribution.

* Interquartile range: range of a distribution encompassing the middle 50% of the observations.

* Variance and standard deviation.
  Variance: mean squared deviation of all the values from the mean.
  Standard deviation: square root of the variance:
 
   Formule 7 (zie bijlage)

  Coefficient of variation: expression for the standard deviation as a percentage of the mean. CV = s / X.
 

-  Measures of shape:

* Skewness: characteristic of a distribution that assesses its symmetry about the mean.

* Kurtosis: measure of the relative peakedness or flatness of the curve defined by the frequency distribution. The kurtosis of a normal distribution is zero. If the kurtosis is positive, the distribution is more peaked than a normal distribution.

Steps in hypothesis Testing:

1.  Formulate Ho and H1.

Null hypothesis: statement in which no difference or effect is expected. If it is not rejected, no changes will be made.

Alternative hypothesis: statement that some difference or effect is expected. Accepting it will lead to changes in opinions or actions.

One-tailed test: test of the null hypothesis where the alternative hypothesis is expressed directionally. For example: Ho: π ≤ 0,40 and H1: π > 0,40 

Two-tailed test: test of the null hypothesis where the alterative hypothesis is not expressed directionally. For example: π = 0,40 and H1: π ≠ 0,40 

2.  Select an appropriate test. Take into consideration how the test statistic is computed and the sampling distribution that the sample statistic follows.
Test statistic: measures how close the sample has come to the null hypothesis.

3.  Choose level of significance, α.
In this two types of errors can occur:

* Type I (alpha error) error: when the sample results lead to the rejection of a null hypothesis that is in fact true.
   Level of significance: probability of making a Type I error.

* Type II (beta error) error: when the sample results lead to the non-rejection of a null hypothesis that is in fact false.
  
The complement (1 – β) of the probability of a Type II error is called: power of
statistical test.
Power of a test: probability of rejecting the null hypothesis when it is in fact false and should be rejected.

4.  Collect Data and calculate Test Statistic.

5.  Determine the Probability (or Critical Value).
P-value: probability of observing a value of the test statistic as extreme as, or more extreme than, the value actually observed, assuming that the null hypothesis is true.

6.  Compare the Probability (or Critical Value).

7.  Make the decision.

8.  Marketing Research Conclusion.

Cross-tabulation: describes 2 or more variables simultaneously and results in tables that reflect the joint distribution of 2 or more variables that have a limited number of categories or distinct values. For example to answer the question: “How many brand-loyal users are males?”.

Contingency table: cross-tabulation table. Contains a cell for every combination of categories of the two variables.

Original Two Variables:
 

-Some Association between the Two Variables:
-> introduce a Third Variable:
    * refined association between the two original variables.
    * no association between the two original variables.
    * no change in the initial pattern.

-No Association between the Two Variables:
-> introduce a Third Variable:
    * some association between the two original variables.
    * no change in the initial pattern.

Chi-square statistic: to test the statistical significance of the observed association in a cross-tabulation. It helps determining whether a systematic association exists between two variables.

Chi-square distribution: skewed distribution whose shape depends solely on the number of degrees of freedom. As this one increases, the chi-square distribution becomes more symmetrical. (Fig 15.8 page 500)

Phi coefficient: measure of the strength of association in the special case of a table with two rows and two columns (2x2 table). Φ = √(X² / n)

Contingency coefficient (C): measure of the strength of association in a table of any size.
C = √ (X² / X² + n)

Cramer’s V: measure of the strength of association used in tables larger than 2x2.

Lambda Coefficient.

Asymmetric lambda: measure of the percentage improvement in predicting the value of the dependent variable, given the value of the independent variable in contingency table analysis. Lambda varies between 0 and 1.

Symmetric lambda: does not make an assumption about which variable is dependent. It measures the overall improvement when prediction is done in both directions.

Other statistics.

Tau b: measures the association between two ordinal-level variables. It makes an adjustment for ties and is most appropriate when the table of variables is square.

Tau c: measures the association between two ordinal-level variables. It makes an adjustment for ties and is most appropriate when the table of variables is not square but a rectangle.

Gamma: measures the association between two ordinal-level variables. It does not make an adjustment for ties.

Hypothesis Testing related to Differences.

àParametric tests: hypothesis-testing procedures that assume that the variables of interest are measured on at least an interval scale.

A parametric test:

T-test: univariate hypothesis test using the t distribution, which is used when the standard deviation is unknown and the sample size is small.

T statistic: assumes that the variable has a symmetric bell-shaped distribution, the mean is (assumed to be) known and the population variance is estimated from the sample.

T distribution: symmetric bell-shaped distribution that is useful for small sample (n < 30) testing, when the mean is known and the population variance is estimated from the sample.

One Sample.

Z-test: univariate hypothesis test using the standard normal distribution.

Two Independent Samples.

Independent samples: two samples that are not experimentally related. The measurement of one sample has no effect on the values of the second sample.

F-test:  statistical test of the equality of the variances of two populations.

F statistic: ratio of two sample variances.

F distribution: frequency distribution that depends upon two sets of degrees of freedom, the one in the numerator and the one in the denominator.

Paired Samples.

Paired samples: In hypothesis testing, the observations are paired so that the two sets of observations relate to the same respondents.

Paired samples t test: test for differences in the means of paired samples.

àNon-parametric tests: assume that the variables are measured on a nominal or ordinal scale.

One Sample.

Kolmogorov-Smirnov (K-S) one-sample test: one-sample nonparametric goodness-of-fit test that compares the cumulative distribution function for a variable with a specified distribution.

Runs test: test of randomness for a dichotomous variable.

Binomial test: goodness-of-fit statistical test for dichotomous variables which tests the goodness of fit of the observed number of observations in each category to the number expected under a specified binomial distribution.

Two Independent Samples.

Mann-Whitney U test: statistical test for a variable measured on a ordinal scale, comparing the difference in the location of two populations based on observations from two independent samples.

Two-sample median test: determines whether two groups are drawn from populations with the same median. This test is not as powerful as the Mann-Whitney U.

Kolmogorov-Smirnov two sample test: determines whether two distributions are the same. It takes into account any differences in the two distributions, including median, dispersion, and skewness.

Paired Samples.

Wilcoxon matched-pairs signed-ranks test: analyzes the differences between the paired observations, taking into account the magnitude of the differences.

Sign test: examines differences in the location of two populations, based on paired observations, that compares only the signs of the differences between pairs of variables without taking into account the ranks.

16. Variance and Covariance analysis.

Analysis of variance (ANOVA): examining the differences between means for two or more populations.

Factors: categorical independent variables. The independent variables must all be categorical (non-metric) to use ANOVA.

Treatment: in ANOVA, a particular combination of factor levels or categories.

One-way analysis of variance: ANOVA technique with only one factor.

N-way analysis of variance: ANOVA model with two or more factors involved.

Analysis of covariance (ANCOVA): advanced analysis of variance procedure where the effects of one or more metric-scaled extraneous variables are removed from the dependent variable before using the ANOVA.

Covariate: metric independent variable used in ANCOVA.

Statistics in One-Way Analysis of Variance:

Eta² (ŋ²): the strength of the effect of X (independent) on Y (dependent). The value varies between 0 and 1.

F statistic: the null hypothesis that the category means are equal in the population is tested by an F statistic based on the ratio of mean square related to X and mean square related to error.

Mean square: sum of squares divided by the appropriate degrees of freedom.

SSbetween/SSx: variation in Y related to the variation in the means of the categories of X. It represents variation between the categories of X or the portion of the sum of squares in Y related to X.

SSwithin/SSerror: variation in Y due to the variation within each of the categories of X. This variation is not accounted for by X.

SSy: total variation in Y.

Conducting One-Way ANOVA:

1.  Identify the dependent and independent variables.
 

2.  Decompose the total variation.
Decomposition of the total variation: in one-way ANOVA, separation of the variation observed in the dependent variable into the variation due to the independent variables + the variation due to error.
 

3.  Measure the effects.
ŋ² = SSx/Ssy     (SSy – SSerror)/Ssy
  

4.  Test the significance:

Formule 8 (zie bijlage)
 

5.  Interpret the results.

The salient assumptions in ANOVA can be summarized in:

1.  Ordinarily, the categories of the independent variables are assumed to be fixed. Inferences are made only to the specific categories considered. This is fixed-effects model.
In the random-effects model the categories or treatments are considered to be random samples from a universe of treatments.
Mixed-effects model results if some treatments are considered fixed and others random.
 

2.  The error term is normally distributed, with a mean of zero and a constant variable. The error is not related to any of the categories of X. The data can be transformed to satisfy the assumption of normality or equal variances.
 

3.  The error terms are uncorrelated.

 N-Way Analysis of Variance.

Interaction: the relationship between two variables, an interaction occurs if the effect of X1 depends on the level of X2, and vice versa.

Multiple ŋ²: strength of the joint effect of two (or more) factors, or the overall effect.

Significance of the overall effect: test that some differences exist between some of the treatment groups.

   Formule 9 (zie bijlage)

Significance of the interaction effect: test of the significance of the interaction between two or more independent variables. 

Significance of the main effect: test of the significance of the main effect for each individual factor.

Analysis of Covariance (ANCOVA).

When examining the differences in the mean values of the dependent variable related to the effect of the controlled independent variable, it is often necessary to take into account the influence of uncontrolled independent variables. So then ANCOVA used.

Interactions.

Ordinal interactions: interaction where the rank order of the effects attributable to one factor does not change across the levels of the second factor.

Disordinal interactions: the change in the rank order of the effects of one factor across the levels of another.

The most commonly used measure in ANCOVA is:

Omega squared (ω²): measure indicating the proportion of the variation in the dependent variable explained by a particular independent variable or factor.

Multiple comparisons.

Contrasts: in ANOVA a method to examine differences between two or more means of the treatment groups.

A priori contrasts: contrasts that are determined before using the analysis, based on the researcher’s theoretical framework.

A posteriori contrasts: contrasts made after the analysis. These are commonly multiple comparison tests.

Multiple comparison test: enable the researcher to construct generalized confidence intervals that can be used to make pairwise comparisons of all treatment means.

Repeated measures analysis of variance: ANCOVA technique used when respondents are exposed to more than one treatment condition and repeated measurements are obtained.

Non-metric Analysis of Variance.

Non-metric analysis of variance: ANOVA technique to examine the difference in the central tendencies of more than two groups when the dependent variable is measured on an ordinal scale.

K-sample median test: non-parametric test used to examine differences between groups when the dependent variable is measured on an ordinal scale.

Kruskal-Wallis one-way analysis of variance: non-metric ANCOVA test used to rank value of each case, not only its location relative to the median.

Multivariate Analysis of Variance.

Multivariate analysis of variance (MANOVA): ANOVA technique using two or more metric dependent variables.

17. Regression and Correlation

Product moment correlation (r): statistic summarizes the strength of association between two metric variables.

Covariance: systematic relationship between two variables in which a change in one implies a corresponding change in the other (COVxy).

Formule 10 (zie bijlage)

Partial Correlation.

Partial correlation coefficient: measure of the association among two variables after controlling or adjusting for the effects of one or more additional variables.

Part correlation coefficient: measure of the correlation between Y and X if the linear effects of the other independent variables have been removed from X but not from Y.

Non-metric Correlation.

Non-metric correlation: correlation measure for two non-metric variables that relies on rankings to compute the correlation. Vary from -1.0 to 1.0.

Regression Analysis.

Regression analysis: statistical procedure for analyzing associative relationships among a metric dependent variable and one or more independent variables.

Bivariate Regression.

Bivariate regression: procedure for drawing a mathematical relationship, in the form of an equation, between a single metric dependent variable and a single metric independent variable.

Statistics associated with Bivariate Regression Analysis.

Bivariate regression model:   

   Formule 11 (zie bijlage)

with:
 

Yi = dependent or criterion variable

Βo = intercept of the line

Β1 = slope of the line

Xi = independent or predictor variable

ei = error term associated with the i-th observation.

Coefficient of determination: strength of association is measured by the coefficient of determination, r ².

Estimated or predicted value: 

  Formule 12 (zie bijlage)

Regression coefficient: the estimated parameter b is usually referred to as the nonstandardized regression coefficient.

Scattergram: plot of the values of two variables for all the cases or observations.

Standard error of estimate: SEE: Formule 13 (zie bijlage)

Standard error: the standard deviation of b: SEb.

Standardized regression coefficient: beta coefficient/weight: is the slope obtained by the regression of Y on X when the data are standardized.

Sum of squared errors: distances of all the points from the regression line are squared and added together to get the sum of squared errors, which is total error.

T-statistic: a t statistic with n-2 degrees of freedom can be used to test the null hypothesis that no linear relationship exists between X and Y, or: Ho = B1, where

Formule 14 (zie bijlage)

Conducting Bivariate Regression Analysis.

Least-squares procedure: technique for fitting a straight line to a scattergram by minimizing the square of the vertical distances of all the points from the line; this procedure is called ordinary least squares (OLS) regression.

Finding the b-values.

Formule 15 (zie bijlage)

Multiple Regression.

Multiple regression: technique that simultaneously develops a mathematical relationship between two or more independent variables and an interval-scaled dependent variable.

Multiple regression model:

Tabel 1 (zie bijlage)

Statistics associated with Multiple Regression:

Adjusted R²: R² is adjusted for the number of independent variables and the sample size to account for diminishing returns.

Coefficient of multiple determination: strength of association in multiple regression is measured by the square of the multiple correlation coefficient (R²) which is also called the coefficient of multiple determination.

F test: used to test the null hypothesis that the coefficient of multiple determination in the population R²pop is zero. This is testing the null hypothesis Ho = B1 = B2 = B3 = … = Bk = 0.

It has an F distribution with k and (n – k – 1) degrees of freedom.

Partial F test: significance of a partial regression coefficient, Bi, of Xi may be tested using an incremental F statistic. The incremental F statistic is based on the increment in the explained sum of squares resulting from the addition of the independent variable Xi to the regression equation after all the other independent variables have been included.

Partial regression coefficient: the partial regression coefficient, b1, denotes the change in the predicted value Ŷ per unit change in Xi when the other independent variables X2 to Xk are constant.

Examination of the Residuals.

Residual: difference between the observed value of Yi and the value predicted by the regression equation Ŷi.

Stepwise Regression.

Stepwise regression: regression procedure in which the predictor variables enter or leave the regression equation one at a time.

The purpose of stepwise regression is to select, from a lot of predictor variables, a small subset of variables that account for most of the variation in the dependent variable.

There are more approaches for stepwise regression:
 

-  Forward inclusion: first there are no predictor variables, then predictor variables are entered (only if they meet certain criteria specified in terms of the F ratio).
 

-  Backward elimination: first all the predictor variables are included, then they are removed one at a time based on the F ratio.
 

-  Stepwise solution: forward inclusion combined with the removal of predictors that no longer meet the specified criterion at each step.

Multicollinearity.

Multicollinearity: state of very high intercorrelations among independent variables.

It can result in problems like:
 

-  Not an precisely estimation of the partial regression coefficients. The standard errors are likely to be high.
 

-  The magnitudes and the signs of the partial regression coefficients may change from sample to sample.
 

-  It becomes hard to assess the relative importance of the independent variable in explaining the variation in the dependent variable.
 

-  Predictor variables may be incorrectly included or removed.
 

In applied marketing research, multicollinearity is valuable to determine the relative importance of the predictors. But because the predictors are correlated, there is no obvious measure of relative importance of the predictors in regression analysis. Although several approaches are commonly used to assess the relative importance of predictor variables:

1.  Statistical significance: If the partial regression coefficient of a variable is not significant, it is judged as unimportant.
 

2.  Square of the simple correlation coefficient: r ² represents the proportion of the variation in the dependent variable explained by the independent variable in a bivariate relationship.
 

3.  Square of the partial correlation coefficient: R² is the coefficient of determination between the dependent variable and the independent variable, controlling for the effects of the other independent variables.
 

4.  Square of the part correlation coefficient: this coefficient shows an increase in R² when a variable is entered into a regression equation that already contains the other independent variables.
 

5.  Measures based on standardized coefficients or beta weights: Beta weights take into account the effect of the other independent variables. These measures become very unreliable as the correlations among the predictor variables increase (multicollinearity increases).
 

6.  Stepwise regression: the order in which the predictors are entered or removed from the regression equation is used to conclude their relative importance.

Cross-Validation.

Cross-validation: test of validity that examines if a model holds on comparable data not used in the original estimation.

Double cross-validation: the sample is split into halves. One half serves as the estimation sample and the other as a validation sample. The roles of both halves are then reversed and the cross-validation process is repeated.

18. Analysis of Discriminant and Logit.

Discriminant analysis: technique for analyzing marketing research data when the criterion/dependent variable is categorical and the predictor/independent variables are interval in nature.

For example dependent variable is choice of brand of computers (brand A, B, C) and the independent variable are ratings of attributes of computers on a 7-point Likert scale.

The objectives of Discriminant analysis:

1.  Develop Discriminant functions: linear combination of independent variables developed by Discriminant analysis that will best discriminate between the categories of the dependent variable.
 

2.  Examine whether significant differences exist among the groups, in terms of the predictor variables.
 

3.  Determine which predictor variables contribute to most of the intergroup differences.
 

4.  Classify cases to one of the groups based on the values of the predictor variables.
 

5.  Evaluate the accuracy of classification.

Two-group Discriminant analysis: the criterion variables has two categories.

Multiple Discriminant analysis: the criterion variables has three or more categories.

Relationship between ANOVA, regression and Discriminant/Logit analysis:
 

Discriminant analysis model:
 

  Formule 16 (zie bijlage)

Statistics Associated with Discriminant Analysis.

Canonical correlation: measures the extent of association between the discriminant scores and the groups. It measures association between the single discriminant function and the set of dummy variables that define the group membership.

Centroid: mean values for the discriminant scores for a particular group.

Classification matrix: confusion/prediction matrix: contains the number of correctly classified and misclassified cases. The correctly classified cases appear on the diagonal, because the predictor and actual groups are the same. The incorrectly classified cases are the off-diagonal elements.

Hit ratio: Sum of the diagonal elements / Total number of cases

Discriminant function coefficients: (unstandardized) multipliers of variables, when the variables are in the original units of measurement.

Discriminant scores: the values of the variables multiplied by the unstandardized coefficients.

Eigenvalue: ratio of between-group to within-group sums of squares, for each discriminant function. Large eigenvalues = superior functions.

F values and their significance: calculated from a one-way ANOVA, with the grouping variable as the categorical independent variable. Each predictor serves as the metric dependent variable in the ANOVA.

Group means and group standard deviations: are computed for each predictor and each group.

Pooled within-group correlation matrix: averaging the separate covariance matrices for all the groups.

Standardized discriminant function coefficients: are used as the multipliers when the variables have been standardized to a mean of 0 and a variance of 1.

Structure correlations: discriminant loadings: represent the simple correlations between the predictors and the discriminant function.

Wilks’ λ: sometimes U statistic called: ratio of the within-group sum of squares to the total sum of squares. Its value varies between 0 and 1. Large values of λ (near 1) implies that group means do not seem to be different. Small values implies that the group means seem to be different.

Steps in conducting Discriminant Analysis:
 

1.  Formulate the problem.
Analysis sample: part of the total sample that is used for estimation of the discriminant function.

Validation (holdout) sample: that part of the total sample used to check the results of the estimation sample.
 

2.  Estimate the discriminant function coefficients.
Direct method: estimating the discriminant function so that all the predictors are included simultaneously.

Stepwise discriminant analysis: the predictors are entered sequentially based on their ability to discriminate between the groups.
 

3.  Determine the significance of the discriminant function.
 

4.  Interpret the results.
Characteristic profile: describing each group in terms of the group means for the predictor variables as aid to interpreting discriminant analysis results.
 

5.  Assess the validity of discriminant analysis.
Hit ratio: percentage of cases correctly classified by discriminant analysis.

These steps can also be followed in the Multiple Discriminant Analysis. Then at interpreting the results can be added:

Territorial map: tool for assessing discriminant analysis results that plots the group membership of each case on a graph.

Stepwise discriminant analysis.

The selection of the stepwise procedure is based on the optimizing criterion adopted.
Mahalanobis procedure: is based on maximizing a generalized measure of the distance between the two closest groups. This allows maximal use of information.

If the dependent variable is binary, and there are several independent metric variables, in addition to two-group discriminant analysis also ordinary least squares (OLS) regression, logit, and the probit models for estimation, can be used.

21. Scaling on Multidimensional level and Conjoint Analysis

Multidimensional scaling (MDS): class of procedures for representing perceptions and preferences of respondents spatially by means of a visual display.
 

The axes of a spatial map denote the psychological bases or underlying dimensions respondents use to form perceptions and preferences for stimuli.

Statistics and terms associated with MDS:

Similarity judgment: ratings on all possible pairs of brands or other stimuli in terms of their similarity using a Likert-type scale.

Preference rankings: rank orderings of the brands or other stimuli from the most preferred to the least preferred.

Stress: lack-of-fit measure; higher values of stress indicate poorer fits.

R-square: squared correlation index indicating the proportion of variance of the optimally scaled data that can be accounted for by the MDS procedure. This is a goodness-of-fit measure.

Spatial map: observes relationships between brands or other stimuli are represented as geometric relationships between points in a multidimensional space.

Coordinates: the positioning of a brand or a stimulus in a spatial map.

Unfolding: representation of both brands and respondents as points in the same space.

Steps in conducting Multidimensional Scaling:

1.  Formulate the problem.
 

2.  Obtain input data.
> perception data: direct approaches: respondents are asked to judge how similar or
   dissimilar the various brands or stimuli are, using their own criteria.

> perception data: derived approaches: attribute-based approaches to collecting
   perception data requiring the respondents to rate the stimuli on the identified
   attributes using semantic differential or Likert scales.
 

3.  Select an MDS procedure.
Nonmetric MDS: type of multidimensional scaling method that assumes that the input data are ordinal.

Metric MDS: assumes that the input data are metric.
 

4.  Decide on the number of dimensions.
The fit of an MDS solution is often assessed by the stress measure. Stress is a lack-of-fit measure, higher values indicate poorer fits.

Guidelines for determining the number of dimensions:
1. A priori knowledge: theory or past research.
2. Interpretability of the spatial map.
3. Elbow criterion: plot of stress versus dimensionality used in MDS. The point
    at which an elbow or a sharp bend occurs indicates an appropriate
    dimensionality.
4. Ease of use. It’s easier to work with two-dimensional maps than with thouse
    involving more dimensions.
5. Statistical approaches.
 

5.  Label the dimensions and interpret the configuration.
 

6.  Assess reliability and validity. Suggestions:
- examine the index of fit, or R-square.
- stress values are indicative of the quality of MDS solutions.
- adding a random error term.
- collection of the input data could be done at two different points in time

Analysis of Preference Data.

Internal analysis of preferences: configuring a spatial map such that it represents both brands or stimuli and respondent points or vectors and is derived solely from the preference data.

External analysis of preferences: configuring a spatial map such that the ideal points or vectors based on preference data are fitted in a spatial map derived from perception data.

Correspondence analysis: MDS technique for scaling qualitative data that scales the rows and columns of the input contingency table in corresponding units so that each can be displayed in the same low-dimensional space.

Conjoint analysis: technique to attempt to determine the relative importance consumers attach to salient attributes and the utilities they attach to the levels of attributes.

Conjoint analysis is used in marketing with several goals:

-  Determine the relative importance of attributes in the consumer choice process.
 

-  Estimate market share of brands that differ in attribute level.
 

-  Determine the composition of the most preferred brand.
 

-  Segment the market based on similarity of preferences for attribute levels.

Statistics and terms associated with conjoint analysis:

Part-worth functions: utility functions describe the utility consumers attach to the levels of each attribute.

Relative importance weights: indicate which attributes are important in influencing consumer choice.

Attribute levels: values assumed by the attributes.

Full profiles: profiles that are constructed in terms of all the attributes by using the attribute levels specified by the design.

Pairwise tables: the respondents evaluate two attributes at a time until all the required pairs of attributes are evaluated.

Cyclical designs: designs to reduce the number of paired comparisons.

Fractional factorial designs: to reduce the number of stimulus profiles to be evaluated in the full profile approach.

Orthogonal arrays: special class of fractional designs that enable the efficient estimation of all main effects.

Internal validity: correlations of the predicted evaluations for the holdout or validation stimuli with those obtained from the respondents.

Steps in conducting the Conjoint Analysis:
 

1.  Formulate the problem.

2.  Construct the stimuli.

3.  Decide on the form of input data.

4.  Select a conjoint analysis procedure.
Conjoint analysis model: mathematical model expressing the fundamental relationship between attributes and utility in conjoint analysis.

5.  Interpret the results.

6.  Assess reliability and validity.
Several procedures for this:
- goodness of fit of the estimated model should be evaluated.
- test-retest reliability can be assessed by obtaining a few replicated judgments
  later in data collection
- evaluations for the holdout or validation stimuli can be predicted by the
  estimated part-worth functions.
- if an aggregate-level analysis is conducted, the estimation sample can be split
  in several ways and conjoint analysis conducted on each subsample.

Hybrid conjoint analysis: form of conjoint analysis which attempts to simplify the data-collection task and estimate selection interactions and main effects.

Year of summary

This summary of Marketing Research, an applied approach (Malhotra) is written in 2014