Glossary of words of terminology used throughout this site, and in general discussions involving economic research.
a b c d e f g h i j k l m n o p q r s t u v w x y z
A
Artificial Intelligence
Attribute
A feature or characteristic. An Attribute has two or more Levels. There is often confusion about the difference between an Attribute and a Level. Think of it this way - Attributes are nouns (colour, size), Levels are adjectives (green, large). Attributes are labels (price), Levels are values ($100). Attributes are the questions consumers ask about a product (what colour is it?, what's the price?), Levels are the answers (green, $100).
Availability Design
A subset of an Experimental Design which dictates whether certain attributes and/or alternatives are visible within a Choice Set.
top
B
Best/Worst (Max Diff)
Best/Worst Scaling is a measurement approach that can measure subjective quantities on ratio scales. The measurement properties of Best/Worst were demonstrated recently by Marley and Louviere (Jl of Mathematical Psych 2005). Sometimes referred to as Maximum Difference or Max/Diff.
top
C
CAPI
Computer Aided Personal Interview. A survey method where respondents are intercepted by an operator with a computer (usually a laptop) in shopping malls or the (suddenly not so) privacy of their homes.
CATI
Computer Aided Telephone Interview. Where survey respondents are contacted by phone and operators read questions from, and enter answers into, a computer terminal. Increasingly ignored in favour of online surveys for populations with good internet penetration.
Choice Model
Any member of the families of statistical models that allow one to explain and predict the choices made by individuals and groups of individuals.
Choice Set
Computer Lab
Where survey respondents complete a computer-based survey on a dedicated network (or even a single computer) set up for that purpose. Commonly used for populations where internet penetration is poor, or for experiments containing rich media (eg video) where high-speed broadband penetration is poor.
Confirmation Bias
Confirmation bias refers to a type of selective thinking whereby one tends to notice and to look for what confirms one's beliefs, and to ignore, not look for, or undervalue the relevance of what contradicts one's beliefs. Researchers are sometimes guilty of confirmation bias by setting up experiments or framing their data in ways that will tend to confirm their hypotheses. They compound the problem by proceeding in ways that avoid dealing with data that would contradict their hypotheses.
Covariates
While a respondent’s choice data describes what they prefer, covariate data describes who they are. Covariates can include Demographics and Psychographics. Covariate questions usually accompanying any choice research. At the very least we will want to know age, sex and income. Covariates are often used to identify Segments.
top
D
Demographics
Demographics are covariates such as age, gender, income, residential postcode. This kind of data is collected in a national census, therefore accurate, recent population-wide statistics are readily available. The composition of any sample rarely matches the entire population (we may have a greater proportion of women, or low income earners), so we use demographics to weight the data and make more accurate predictions about the entire population.
Design Coding
When data has design coding applied, the first level will be 0 (zero), not 1.
Discrete Choice Experiment (DCE)
A DCE provides the data needed to create a Discrete Choice Model. While it is theoretically possible to attempt to construct a Discrete Choice Model using historical data, it is highly unlikely in practice, since historical data do not comply with an Experimental Design. To put it another way - the features and price of a product throughout a period of history are likely to follow a narrow band. Without varying features and price widely across the space of all possible permutations (as an Experimental Design does) it is impossible to make accurate extrapolations and interpolations about the utility of a useful range of permutations. Some DCEs are small, involving a few comparison sets; others are large, involving hundreds of comparison sets. The size of a DCE depends on the number of features and their levels, and cannot be specified a priori without knowing the exact number of attributes, and the exact number of levels of each attribute. Once you have specified attributes and levels, then created an appropriate Experimental Design, you have a basic DCE. See also Stated Preference.
Discrete choice modelling (DCM)
Discrete Choice Modelling means using data collected from a Discrete Choice Experiment to create a Choice Model. The model explains and predicts the choice behaviour of a population or individual - eg. how much they are likely to pay for a service, what the market share of a new product will be, whether they will vote for a set of policies, whether they will accept a new job, how many staff will resign in the next year, how many potential customers will redeem a coupon, what percentage of doctors will diagnose a condition correctly... the number of applications are endless.
top
E
EBA Model
Elasticity
Elasticity is used to measure the exact amount of volume shift responding to the variable under scrutiny.
Experimental Design
Expert Focus Groups
top
F
Factor Analysis
Factor analysis of data identifies the underlying factors that explain the correlation among a set of variables. It is used to identify a new, smaller use of uncorrelated variables to replace the original set of correlated variables. It also identifies a smaller set of salient variables for use in subsequent research. It is used in various types of research including product research, customer satisfaction, market segmentation and consumer profiling.
Fieldwork
Fieldwork can be conducted by observation, surveys (such as face to face interviews, telephone interviews and web interviews) or experiments. It is the basic term for the live collection of primary data from external sources. Fieldwork is either co-ordinated by an in-house fieldwork department within a market research agency or an external fieldwork company. Once fieldwork has been conducted data processing is usually the next step.
Focus Groups (See also Expert Focus Groups)
In a focus group, respondents (normally between 8-10 people) are gathered together in order to gauge their responses to specific stimuli. Groups are guided by a research moderator who often uses a topic guide to control the discussion to ensure it meets the initial research objectives. The data generated is probably most applicable to exploratory work. The technique falls under the broad category of qualitative research.
top
G
Game Theory
H
Heterogeneity
top
I
Incentive
This is the payment made to a respondent in return for their participation in a research project and depending on the complexity and duration of the research, the incentive will vary.
Information Conditioning (also Information Acceleration)
Interaction
top
L
Level
One of the two or more specified values of an Attribute. For a full definition of the both Attributes and Levels and their relationship, see also Attribute.
Likert Scale
A typical question using a Likert scale (pronounced 'lick-ert') might make a statement, then ask the respondents to indicate their degree of agreement on a categorical scale. Typically a five-point scale is used.
Longitudinal Study
A research study which follows a group of subjects over an extended period of time, often several years.
top
M
Max/Diff
See Best/Worst
Mean
A simple average value of a set of data.
Median
Another form of averaging data, the median value is the middle value of a set of data when the data is arranged in terms of magnitude (e.g. in the set 1,2,3,4,5,6,7 - four would be the median).
Mixed Logit Mixed Logit is the most theoretically robust kind of
Discrete Choice Model. It will account for Preference Heterogeneity. Mixed Logit assumes there are no fixed utilities for attribute-levels across the entire population, i.e, we can’t say the population will pay $20 more for the blue car – that utility parameter is random, and some people will people $50 more or $10 less. Mixed Logit models (also called Random Parameters models) will produce a distribution that predicts how the utility of ‘blue’ changes across the population.
Monadic Scaling
This is where a respondent is asked to rate an item on some form of scale without comparison to another product or service.
Multinomial Logit (MNL)The first, most elegant, form of logit model, developed by Dan McFadden. MNL is also the easiest to model, and to specify an experiment for. The assumptions it rests on can be challenged in many real-life choice situations, however those working in the applied science of
DCM will say that MNL models can still provide extremely accurate and pertinent data, even where assumptions are probably violated. In the case of violations, the superior and more accurate models will be
Nested Logit or
Mixed Logit.
top
N
Nested Logit
In nested logit models we assume that some alternatives are nested under a category that is important to the respondent, features in their decision making, contributes to the utility of the alternatives (negatively or positively) and therefore needs to be estimated in the model. For example, in a choice between travelling to work on a tollway, surface roads, by bus or by train, we can reasonably assume that the first two options are nested under the category ‘Car’ and the last two are nested under the category ‘Public Transport’. We have to identify these categories, and measure their effect – this will give us a better model than if we assume the four options are independent unrelated alternatives.
Nominal Scale
A scale of measurement in which numbers don't represent value, instead their use is to label or catagorise an object or an event.
top
P
Panels
A panel is a long-standing sample that is retained by a market research agency from which data can be attained. It is most useful for continuous research whereby the same set of respondents are used on a continuous basis over time.
Probit
Psychographics
A loose term describing covariates that are more informal, subjective, temporary, unstable, psychological, behavioural or diagnostic, compared to demographics. Psychographics include attitudes, values, views on life, responses to statements, favourite movies, and scales designed to diagnose a certain psychological condition or disposition.
top
Q
Qualitative
Qualitative research is in-depth research performed on a small scale to provide detailed in depth results and data. Qualitative research can be performed over the phone, via group discussions or through one-to-one interviews. Discussions are normally aided by using a 'topic guide' - this outlines the basic structure of the interview/group and indicates the general direction in which the interview/group should be led. Questions are of open nature as opposed to closed questions which are used in quantitative research.
Quantitative
Quantitative research (quant) is performed on a far larger scale compared with qualitative research (in terms of the sample size) and helps to provide accurate statistical data from which conclusions can be drawn. Questions tend to be closed as opposed to open
top
R
Random Utility Theory (RUT)
Under Random Utility Theory (RUT), the alternative with the highest utility is chosen, just like Classical economics. However, Random Utility Theory posits that utility has a deterministic, systematic component, and a ‘stochastic’ component which captures uncertainty, error, unobserved attributes, taste variations. In other words, it accounts for the fact that humans simply change their minds, make mistakes, change their tastes from time-to-time, etc. It also, handily, accounts for the fact that the analysts observing choice behaviour are also human and can make measurement errors, fail to observe certain attributes, etc. The mathematical expression of RUT is like the E=mC2 of discrete choice models. It is the basis for all the maths and theory. RUT is also the reason utilities in DCM are expressed as a probability, and the probability that an individual in the population will choose an alternative is the market share of that alternative, when applied to the entire population.
Respondent
A respondent is the person whose views and opinions are required by the researcher.
Response Bias
Response bias is the term for inaccurate response data which could have been caused by a number of factors such as monotony, tiredness, aiming to please etc.
Revealed Preference
see Stated Preference
top
S
Sample
A quota sample is one which has been selected to represent the population, sometimes incorporating certain control characteristics. In research terms, a sample refers to a group of interviewees or respondents who are chosen to represent the population as a whole. The sample provides the data within the market research project.
Scenario
see Choice Set
Segments / Segmentation
The subsection of the general population which we are studying, modelling, and/or generating predictions for. Segments can be specified prior to a research project, or identified in the course of analysis. Prior specification occurs when, for example, we are studying a beer brand, specify a segment of “males 18-35 years old”, or “beer drinkers”, and only collect data from this segment. Analytical identification occurs when we collect covariate data and use it to model segments and compare the segment models to the general population model. For example, in our Space Tourism study, we found that people with higher incomes, and people who participated in dangerous sports, showed a higher demand for space travel, and displayed strong preferences for particular packages, compared to the general population. Sometimes the factors that make up a segment of interest are arrived at by educated guess, often they are derived by using Latent Class Models.
Stated Preference
What human beings say they will do, as opposed to what they are observed to actually do. A choice an individual makes between two products in a survey is a Stated Preference. The same choice, between the same goods, in a supermarket or on a website, where the individual actually buys one or the other (or neither) is a Revealed Preference. DCEs are usually Stated Preference (i.e, surveys), but can be Revealed Preference in the unusual case where we can actually offer for sale all of the product permutations dictated by the Experimental Design. This could mean, for instance, that we are prepared to sell some products below cost, and also prepared to deal with the negative sentiment of customers offered unusually high prices. While experimental RP data are useful, historical RP data are of no practical use for Discrete Choice Modelling - why? - see Discrete Choice Experiment.
top
T
Target Population
The basic term for the population that is being studied.
Test Market
A test market is used as a trial market for a new product or service.
top
V
Validation
The automatic process by which a computer-based task will not allow a nonsensical response.
top
