An experiment is said to possess internal validity in so far as it controls the effects of all non-manipulated variables so that the differences among experimental groups on the dependent variable can be treated as valid effects of the manipulations. An experiment is internally valid so far as the observed differences among the treatment groups are valid effects of the manipulations. In other words, internal validity removes causal uncertainty.

Several factors affect the internal validity of the experiments.

Selection bias The participants in the experimental groups may be different from those of other groups. The effects then could be attributed to the differences in the groups, rather than the manipulation of the causal variable.

Treatment effects The very fact that it is an experiment tend to affect the dependent variable. The subjects or participants tend to fall in line with the experimental thinking e.g. people tend to choose a cover design that they think is desirable. Participants may also tend to look more favourably.

Testing effects When a dependent variable is being measured after manipulation, the testing procedure tends to affect the results. It is because the pre-test tends to make us fall in line with the desired outcome at the time of post-test.

History effects These happen when an outside event affects the dependent variable during one experiment, e.g. competitive action.

Maturation Effects In this there is some change in the participants which is unrelated to the manipulation. It is this change that affects the dependent variable.

Mortality Effects Some participants may not last till the end of the experiment. Desertions do affect the observed effects.

The following table summarises the above discussion.

Threats to Internal Validity

Threat Reason

Selection bras O Differences in the participants.

Treatment Effects O The very fact that it is an experiment

O The desire to look favourable

O The desire to fall in line with the researcher’s thinking.

Testing Effects O Interaction with the measurement procedur

History Effects O External environment changes

Maturation effects O Internal changes in the participants.

Mortality Effects O Some participants desert the experiment

External Validity

An experiment is externally valid so far as the effects occuring are closer to those in an actual market situation. It is also called generalizability as the effects are generalized to the market place. In the book cover experiment, the two designs sent to different booksellers are high on external validity. Test marketing is also high on enternal validity. When book covers are compared in office by the employees of the publishing house, the experiment is low on external validity. Here book covers are not tested under realistic market conditions. The employees are not representative of buyers of the books.

Selection bias The participants in the experimental groups may be different from those of other groups. The effects then could be attributed to the differences in the groups, rather than the manipulation of the causal variable.

Treatment effects The very fact that it is an experiment tend to affect the dependent variable. The subjects or participants tend to fall in the line with the experimental thinking, e.g. people tend to choose a cover design that they think is desirable. Participants may also try to look more favourably.

Testing effects When a dependent variable is being measured after manipulation, the testing procedure tends to affect the results. It is because the pre-test is in line with the desired outcome at the time of post-test.

History effects These happen when an outside event affects the dependent variable during one experiment, e.g. competitive action.

Maturation effects In this there is some change in the participants which is unrelated to one manipulation. It is this change that affects the dependent variable.

Marketing effects Some participants may not last till the end of the experiment. Desertions do affect the observed effects.

Broadly speaking, an experiment occurs when some action leads to some results. In a way, we can say that if prices are reduced, we can study its effect by observing whether sales go up. It is a simple experiment where sales will be observed before the price reduction and after the price reduction. It is before-and-after comparison on account of manipulation of price. In a simple experiment, the design remains confined to a single participant or group of participants and is single-group pre-test vs. post-test design.

Some thought must be given to interpretation of results in simple experiments. Suppose we reduce the price of text-books by 10% and the sales of the books go down by 20 per cent. What does it mean? Is it necessary to raise the price? Maybe, the professors have suggested the use of guide-books and help-books. Maybe, someone else has reduced the price by 15 per cent. What if we conduct the experiments after examinations?

This is where simple experiment’s results are vulnerable to effects of many factors other than manipulation. We can observe the change in results, i.e. the change in sales of a dependent variable. But we cannot attribute these changes to manipulation alone. Non-manipulated activities also may affect the results. An experiment is useful only so far as it can control the effects of other events which may affect the dependent variable, so as to make changes attributable to manipulation alone. One way out to isolate non-manipulated variables in experiments is to do lab research. For instance, we can control price of a competing book-store in a lab experiment. But lab experiments do not reflect reality. Besides, they are also not free from many extraneous factors.

The other way is to set up comparison group in the experiment to control the effects of non-manipulated variables. A bookstore can measure the effect of reduced prices on sales, by reducing prices in one group of markets and keeping them constant in another group. The group where prices are lowered is called the experimental group and the group where they are kept constant is called the control group. If the groups chosen are to begin with on par, the effect of manipulation can be measured by comparing the groups. An extraneous factor affects both the groups and cancels out its effect.

In an experiment, a causal variable is manipulated actively to measure how it has affected the dependent variable of interest. There can be one or more causal variable and dependent variable.

To illustrate,

O A publishing company prints several cover designs for its advertising text book, and asks the students which cover design do they like the best. This experiment tests the effects on cover design on preference.

O The publishing company prints two different cover designs and sends the advertising book to different book sellers. It then measures the sales for each design. In this experiment, we measure how a cover design affects the sales.

O A butter producer can ask its customers to rate the butter on its taste, spreadability and thickness. The experiment measures the effects of various attributes on product ratings.

O A music system can have four different price levels, four different brand names, two different power sources (mains plus battery) and two different sources of music — tape and radio. The consumers are to rank these sixty four combinations from most preferred to least preferred. The attributes and their effect on preference was measured.

The experimental variables in all the above examples have certain outcomes, i.e sales, preference which are dependent variables. Outcomes are produced since the experimental variables are manipulated. It is this active manipulation which makes the studies an experiment.

Experiments can take many forms. They can be conducted at homes, in shops, in marketplace. The manipulations can be physical or hypothetical. There can be any number of manipulations. There can be a measurement of any number of dependent variables. The unit of measurement can be an individual, household, companies, markets etc.

O Marketers of consumer non-durables and fast moving consumer goods.

Companies producing fast moving consumer goods such as Lever, P&G, Godrej are the main users of MR.

O Marketers of consumer durables

Companies producing fridges, washing machines, TVs, sewing machines, fans, ACs, ovens etc, are next to FMCG marketers in the use of MR.

O Industrial marketers

There is need for MR in business-to-business transactions of equipment, plant, machinery etc. The research method employed is different. FMCG companies can do research by mail or telephone surveys but an industrial marketer has to rely upon personal survey. An FMCG marketer conducts a tracking studies to monitor the brand, but an industrial marketer need not do so.

0 Service companies

Banks, insurance companies, hotels, airlines also use MR. They are more interested in customer satisfaction research.

O Non-profit organisations (NPOS)

There are social service organisations like Red Cross, Rotary, Giants, CRY etc. who conduct informal and inexpensive research.

O Retailers

Retailers rely upon the marketing organisations for research findings. However, organised mass retailing is doing research in customer traffic, sales promotion methods, and customer satisfaction.

O Ad agencies

Finally, much of MR was done by the agencies. As MR is spreading in industry, it relies less on agencies. MR is done by agencies for smaller clients who do not have a formal MR arrangement. However, agencies have traditionally undertaken research on advertising effectiveness and media planning. Most clients rely upon the agencies for this expertise.

O Media companies

Media undertakes readership profile or viewership profile studies. They use these studies internally to strengthen the media. Externally, these are their tools in selling space or time. But media provided data tends to show media in favourable light. While using the data, there should be a counter-check by comparing data from other sources.

O Government agencies

Government agencies make use of MR data in population control, health planning, economic planning, resourse allocation etc.

Syndicated research services

When a research firm sells the research findings to a number of clients, it is called syndicated research. The data tracked by the syndicate becomes input into the marketing information system of the client. ORG Retail Audit is an example of syndicated research. Syndicated research is a continuous tracking of market.

Job Shops

These research firms provide customised research studies to their clients. These firms have to plan, design and execute a research project for the client. There are companies like ORG-MARG providing such customised research.

Government Agencies

There are organisations like Agricultural Prices Commission, Indian Statistical Organisation, Institute of Population Studies, Indian Institude of Public Administration, Indian Institute of Forest Management, Reserve Bank of India etc. which provide research data to the organisations. Mostly these institutions generate data which can be used as secondary data by the organisations.

In-house Research

Most organisations prefer to buy research studies from outside, rather than conduct them in-house. But there are exceptions. An organisation generates a lot of data in its account and sales department. It becomes input in the marketing information system. Some companies undertake research project also.

Advertising Agencies

All good ad agencies do brand research and advertising research these days. All agencies have a department of media which undertakes media research.

1 Business and Corporate Research

O Market share studies

O Industry characteristics

O Market characteristics

O Market trends

2. Product Research

O Concept development and testing

O Brand research

O Packaging research

3. Distribution Research

O Channel performance studies

O Warehouse location studies

4. Pricing Research

O Cost analysis

O Demand analysis (Sales forecasts and potential)

O Competitive pricing analysis

5. Promotion Research

O Media research

O Advertising effectiveness

O Motivation research

O Salesforce territories, Compensation etc.

6. Consumer Behaviour

O Attitudes

O Satisfaction

O Awareness

O Segmentation.

Some of the common questions used in marketing analysis are given below:

Pertaining to Customers

O What is the size of the actual market?

O What is the size of the potential market?

O What is the growth rate of the market?

O What is the decision making process while buying?

O Why do the customer buy?

O How do they use the product?

O Do they continue to buy the same product?

O What are the market segments and how big they are?

Pertaining to Competition

O What is our market share vs. competitors?

O What is the sales forecast for us and for our competitors?

O What is the brand awareness visa vis competitors?

O What is the brand perception of our product as compared to competitive products?

O How frequently do they buy our product as against competitive products?

O What is the satisfaction level with our products as against competitive products?

O What is competitive strategy?

Pertaining to Operations

O How good is our distribution?

O How good is our promotion?

O How good are our sales people?

O How good is our pricing?

O What would be the reaction of customers to product changes ?

O What are the chances of new product adoption?

Pertaining to Environment

O How the technological changes will affect us?

O How the media scene would affect us?

O How the social changes would affect us?

O How the government would affect us?

O How the economic environment would affect us ?

Marketing research is defined as all activities that provide information to guide marketing decisions. — Seymour Sudman and Edward Blair

It is a very comprehensive definition. It equates marketing research to information gathering. This information gathering facilitates the strategic and operational decisions about

O target markets

O competitive strategies

O marketing mix consisting of product, price, place (disribution ), or promotion

There are so many methods to do marketing research. It assumes many forms. But the main point is the reason behind marketing research. The content of marketing research is OK. But its application is important — facilitation of the decision making process.

The research methodology is common to those used elsewhere, say economics, sociology, psychology.

Common Issues in MR

MR is basically geared to the 4 Cs — Customers, Competitors, Company and Climate. It seeks to answer questions such as:

O Who are our customers? Where are they? What do they buy? Why do they buy? When do they buy? How much quantity do they buy?

O Who are competing with us? How strong are they? How do they compare with us?

O What are our strengths and weaknesses? How good are our marketing programmes?

O How the environment is changing in terms of technology, sociology and economy?

Data mining is highly valuable as a tool to assess the consumer behaviour. It is applied to develop the products, to price them and to promote them. Banking, finance and insurance (BFI) use data mining traditionally. The next extensive use is in the retail sector. The technique can be applied in the management of the supply chain. Supply chain has to integrate the supply and demand chain end-to-end. In this investigation, there is always an element of uncertainty leading to a mismatch between the demand and supply. Organisations take recourse to software to balance these. Uncertainty occurs in demand, supply and the process matching these two. We have to predict this uncertainty accurately.

In the simplest form, the supply chain looks like

Supplier— Supplier— Producer—- Distributor— Retail– Customer

Marketing flow

Information flow

There are two components. The suppliers on the one end. The distributors and the customers on the other end. In between, there is production process. Thus the supply and demand — both require forecasting to lessen the uncertainty. The method of forecasting conventionally used have limitations. A model can extrapolate the past data into future, based on known parameters and using statistical techniques. The limitations are inaccuracies in forecasting, the number of parameters used and the coefficients of these parameters. If data mining is used, these limitations can be overcome. Here the model is built and rebuilt repeatedly to approximate the reality. Traditional methods use imagination. In data mining the minor effects of some parameters can also be detected which imagination possibly could not.

Data mining is an interdisciplinary subject. Many definitions are possible. As we know, gold is mined from rocks or sand. This is called gold mining and not rock mining. As data mining results into knowledge, it could have been called knowledge mining from data. But this is too long. And if we shorten it to knowledge mining, it undermines the importance of a large volume of data. Even then mining retains the flavour of the process of extracting knowledge from a large amount of data. Some other terms are used to denote data mining — knowledge mining from data, knowledge extraction, data analysis, data archaeology and data dredging. Datamining is used synonymously with knowledge discovery from data (KDD).

Datamining ferrets out a pattern or relationship in the data. The data is subjected to statistical analysis and modeling techniques. Data mining is associated with knowledge discovery. There is a difference between datamining and online analytical data processing — OLAP. In OLAP, a hypothesis is tested by the user. However, auto mining itself generates the hypothesis. The financial implications are considered before applying the results of the detected pattern. OLAP answers myriad of questions. In the initial stages, OLAP is useful to understand data. The important variables are identified. The exceptions are noted. The interactions are discovered. These operations enhance our understanding of data. Datamining process has four stages:

I. Data Warehousing

Here data is managed for decision support. Data is collected, cleaned and converted from systems and other third-party sources. This constitutes the data warehouse. It is the foundation for data mining. The quality of data matters for good results. Data from legacy system is transferred, cleaned and analysed by storing it on central repository and making it available. The conversion of data must free it from irregularities.

As an alternative, data mart is used. It is a functional or departmental repository. The data is sourced from the systems critical to the unit possessing data mart and from select external agencies. These could be individual components.

The problem is that of inconsistency between the architecture of different data marts and data warehouse.

Though a data ware house is not a precondition for datamining, it is better to use it for larger data bases. A data warehouse partitions an online transaction processing system — OLTP of an organisation from its decision support system – DSSs. The data warehouse data is subjected to analysis by the executives. The data then becomes information that aids decision making. The data is classified subject-wise. The data is used for comparisons, trends and forecasting.

2 Data Mining Tools

Data extracted from a data warehouse generates a predictive model/ rule set. We can use algos to do this. The customers can be classified on the basis of characteristics and can be subjected to neural networks or decision trees. The segmentation analysis or clustering can also be subjected neural networks or decision trees. The probability of a customer preferring one product and also preferring another is a problem of association and sequencing. Here statistical techniques or rule induction can be applied. In forecasting, regression can be used. There should be a team effort.

3 Predictive Modelling

We have to optimise by selecting either one or a combination of the predictive models. These models could be developed statistically or may be derived from datamining. These could be done by the modellers or sourced from the external agencies. The predictive models are aggregated. The techniques can be blended sequentially or concurrently.

In the last few decades staticians and computer scientists have produced a dazzling arsenal of extremely powerful tools to help managers translate data into business decisions. Managers have to pick a golden model — one that is neither too complicated nor too simple. The simplest model can be run on Excel. The most complicated is a full-blown Hidden Markov Model — HMM which generally requires the use of specialised programming languages and takes much longer to run.

4. Predictive Scoring

Here the scoring is done for operational data. Banking customers who could not keep the minimum balance in their accounts could be extracted through datamining. These customer profiles could be used for business purposes.