Section 7. Glossary

Adaptive experiment:
This is a "quasi-experimental" research methodology that involves before and after measures, a control group, and non-random assignment of human subjects. Data are gathered before the independent variables are introduced, but the final form is not usually known until after the independent variables have been introduced and the "after" data has been collected (Jenkins, 1985).

Archival research:
This methodology is primarily concerned with the examination of historical documents. Secondarily, it is concerned with any recorded data. All data are examined ex-post-facto by the researcher (Jenkins, 1985).

ANOVA:
Univariate analysis of variance (ANOVA) is a statistical technique to determine, on the basis of one dependent measure, where samples come from populations with equal means. ANOVA is robust to violations of this assumption, however (Lindman, 1974). Univariate analysis of variance employs one dependent measure, whereas multivariate analysis of variance compares samples based on two or more dependent variables (Hair et al., 1995).

ANCOVA:
Analysis of covariance (ANCOVA) is a form of analysis of variance that tests the significance of the differences among means of experimental groups after taking into account initial differences among the groups and the correlation of the initial measures and the dependent variable measures. The measure used as a control variable - the pretest or pertinent variable - is called a covariate (Kerlinger, 1986).

Canonical correlation:
With canonical analysis the objective is to correlate simultaneously several metric dependent variables and several metric independent variables. The underlying principle is to develop a linear combination of each set of variables (both independent and dependent) to maximize the correlation between the two sets (Hair et al., 1995).

Case studies:
Case studies involve the intense examination of a small number of entities by the researcher, where neither independent variables are manipulated nor are confounding variables controlled. Like field studies, case studies typically utilize questionnaires, coded interviews, or systematic observation as their preferred techniques for gathering data (Boudreau et al., 2001).

Cluster analysis:
Cluster analysis is an analytical technique for developing meaningful sub-groups of individuals or objects. Specifically, the objective is to classify a sample of entities (individuals or objects) into a smaller number of mutually exclusive groups based on the similarities among the entities (Hair et al., 1995).

Common Methods Bias:
Common methods bias occurs when the instruments the researcher employs enter into or affect the scores or measures that are being gathered. This is also known as a methodological artifact. In Straub, Gefen, and Boudreau (2004), there is a relevant example from the Technology Acceptance Model (TAM) research stream. The first question that a researcher asks is whether someone perceives the system to be useful. The next question is: "Do you use the system?" To avoid cognitive dissonance, subjects would be likely to correlate their second question highly with their first question. So we are not getting at the "true" score from the user because of common methods bias. Campbell and Fiske (1959) argue that one needs comparative methods to know whether there is such bias. Recent work suggests other ways to test for common methods bias, even when one lacks a second method (Podsakoff et al., 2003).

Conjoint analysis:
Conjoint analysis is an emerging dependence technique that has brought new sophistication to the evaluation of objects, whether they are new products, services, or ideas. The most direct application is in new product or service development, allowing for the evaluation of the complex products while maintaining a realistic decision context for the respondent (Hair et al., 1995).

Convergent validity:
Convergent validity according to Campbell and Fiske (1959) is when, in the presence of other scale items for other constructs, the scale items in a given construct move in the same direction (for reflective measures) and, thus, highly correlate. In a factor analysis, we would expect to see such items loading together on one factor (and not cross-loading on another construct altogether). Convergent validity differs from reliability in that tests of reliability include only the scale items for a single construct and they are not being compared to other constructs. See also discriminant validity, which is the complement of convergent validity and together form the construct validity of an instrument.

Correspondence analysis:
Correspondence analysis is a recently developed interdependence technique that facilitates both dimensional reduction of object ratings (e.g., products, persons, etc.) on a set of attributes and the perceptual mapping of objects relative to these attributes (Hair et al., 1995).

Dependent Variable:
Presumed effect of, or response to, a change in the independent variable(s).

Discriminant validity:
Discriminant validity according to Campbell and Fiske (1959) is when, in the presence of other scale items for other constructs, the scale items in constructs being compared do not move in the same direction (for reflective measures) and, thus, do not highly correlate. If the lack of correlation is as expected by the formulation of these constructs, then we can say that we have established discriminant validity. In a factor analysis, we would, for instance, see unrelated items loading on different factors. See also convergent validity, which is the complement of discriminant validity and together form the construct validity of an instrument.

Experimental Simulation:
This methodology employs a closed simulation model to mirror a segment of the "real world". Human subjects are exposed to this model and their responses are recorded. Thee researcher completely determines the nature and timing of the experimental events (Jenkins, 1985).

Factor analysis:
Factor analysis, including variations such as component analysis and common factor analysis, is a statistical approach that can be used to analyze interrelationships among a large number of variables and to explain these variables in terms of their common underlying dimensions (factors). The objective is to find a way of condensing the information contained in a number of original variables into a smaller set of variates (factors) with a minimum loss of information (Hair et al., 1995).

Field experiments:
Field experiments involve the experimental manipulation of one or more variables within a naturally occurring system and subsequent measurement of the impact of the manipulation on one or more dependent variables (Boudreau et al., 2001).

Field studies:
Field studies are non-experimental inquiries occurring in natural systems. Researchers using field studies cannot manipulate independent variables or control the influence of confounding variables (Boudreau et al., 2001).

Free simulation experiment:
This methodology is similar to experimental simulation, in that with both methodologies the researcher designs a closed setting to mirror the "real world" and measures the response of human subjects as they interact within the system. However, with this methodology, events and their timing are determined by both the researcher and the behavior of the human subject (Jenkins, 1985; Fromkin and Streufert, 1976).

Group feedback:
Employing this methodology, groups of human subjects complete an objective instrument for testing of the researcher's initial hypothesis. Following the statistical analysis of the collected data, the data and the analysis are discussed with the subject group to obtain their subjective evaluation. The intent is to achieve a deeper analysis than that afforded by the statistical analysis alone. This methodology allows a re-evaluation of the original hypotheses (Jenkins, 1985).

Hotelling's T²:
Test to assess the statistical significance of the difference between two sets of sample means. It is a special case of MANOVA used with two groups or levels of a treatment variable (Hair et al., 1995).

Independent Variable:
Presumed cause of any change in a response or dependent variable(s).

Interviews:
An interview is a two-way purposeful conversation initiated by an interviewer to obtain information that is relevant to some research purpose. The participants are typically strangers and the topics and pattern of discussion are dictated by the interviewer. Interviewing is perhaps the most ubiquitous method of obtaining information from people. Interviews are ordinarily quite direct and a great deal of information is generally got from respondents by direct questioning (Emory, 1980; Kerlinger, 1980).

Laboratory experiments:
Laboratory experiments take place in a setting especially created by the researcher for the investigation of the phenomenon. With this research method, the researcher has control over the independent variable(s) and the random assignment of research participants to various treatment and non-treatment conditions (Boudreau et al., 2001).

Linear probability models:
In this technique, one or more independent variables are used to predict a single dependent variable. Linear probability models accommodate all types of independent variables (metric and non-metric) and do not require the assumption of multivariate normality (Hair et al., 1995).

Linear regression:
A linear regression uses the method of least squares to determine the best equation describing a set of x and y data points.

LISREL:
A procedure for the analysis of LInear Structural RELations among one or more sets of variables and variates. It examines the covariance structures of the variables and variates included in the model under consideration. LISREL permits both confirmatory factory analysis and the analysis of path models with multiple sets of data in a simultaneous analysis.

Loading (Factor Loading):
Weighting which reflect the correlation between the original variables and derived factors. Squared factor loadings are the percent of variance in an observed item that is explained by its factor.

LOGIT:
Logit analysis is a special form of regression in which the criterion variable is a non-metric, dichotomous (binary) variable. While differences exist in some aspects, the general manner of interpretation is quite similar to linear regression (Hair et al., 1995).

Math or analytical modeling:
This methodology models the "real world" and states the results as mathematical equations. It is a closed deterministic system in which all of the independent and dependent variables are known and included in the model. Intervening variables simply are not possible and no human subject is required (Jenkins, 1985).

MTMM:
Multitrait-multimethod uses a matrix of correlations representing all possible relationships between a set of constructs, each measured by the same set of methods (Campbell and Fiske, 1959). This matrix is one of many methods that can be used to evaluate construct validity by demonstrating both convergent and discriminant validity.

Multidimensional scaling:
In multidimensional scaling, the objective is to transform consumer judgments of similarity or preference (e.g., preference for stores or brands) into distances represented in multidimensional space. If objects A and B are judged by respondents as being the most similar compared with all other possible pairs of objects, multidimensional scaling techniques will position objects A and B in such a way that the distance between them in the multidimensional space is smaller than the distance between any other two pairs of objects. The resulting perceptual maps show the relative positioning of all objects, but additional analysis is needed to assess which attributes predict the position of each object (Hair et al., 1995).

Multiple regression:
Multiple regression is the appropriate method of analysis when the research problem involves a single metric dependent variable presumed to be related to one or more metric independent variables. The objective of multiple regression analysis is to predict the changes in the dependent variable in response to the changes in the several independent variables (Hair et al., 1995).

Multiple discriminant analysis:
If the single dependent variable is dichotomous (e.g., male-female) or multichotomous (e.g., high-medium-low) and therefore non-metric, the multivariate technique of multiple discriminant analysis (MDA) is appropriate. As with multiple regression, the independent variables are assumed to be metric (Hair et al., 1995).

Multivariate analysis of variance (MANOVA):
Multivariate analysis of variance (MANOVA) is a statistical technique that can be used to simultaneously explore the relationship between several categorical independent variables (usually referred to as treatments) and two or more metric dependent variables. As such, it represents an extension of univariate analysis of variance (ANOVA). MANOVA is useful when the researcher designs an experimental situation (manipulation of several non-metric treatment variables) to test hypotheses concerning the variance in group responses on two or more metric dependent variables (Hair et al., 1995).

Normal Distribution:
A normal distribution resembles the shape of a bell curve with many observations at the mean and a continuously decreasing number of observations as the distance from the mean increases. This symmetrical continuous rate is determined by the variance. A normal distribution, also known as a student distribution and as a bell-curve distribution, is probably the most important type of distribution in behavioral sciences and is the underlying assumption of many of the statistical techniques discussed here. Data generated from a large random sample will, according to the central limit theorem, approximate such a distribution. Graphically, a normal distribution of X will resemble an elastic band held between two horizontal points with a weight at its center, the center being analogous to the mean.

Multinormal distribution:
Also known as a Joint Normal Distribution and as a Multivariate Normal Distribution, occurs when every polynomial combination of items itself has a Normal Distribution. For example, in Linear Regression the dependent variable Y may be the polynomial combination of aX1+bX2+e, where it is assumed that X1 and X2 each has a normal distribution. Multinormal distribution occurs when also the polynomial expression aX1+bX2 itself has a normal distribution. Graphically, a multinormal distribution of X1 and X2 will resemble a sheet of paper with a weight at its center, the center being analogous to the mean of the joint distribution.

Objective Tests:
It is a systematic procedure in which the individuals tested are presented with a set of constructed stimuli to which they respond. These responses enable the tester to assign numerals or sets of numerals to the testees, from which inferences can be made about the testees' possession of whatever the test is supposed to measure (Kerlinger, 1986).

Observation:
Observation means looking at people and listening to them talk. One can infer the meaning, characteristics, motivations, feelings and intentions of others on the basis of observations (Kerlinger, 1986).

Opinion research:
The objective of this methodology is to gather data on attitudes, opinions, impressions and beliefs of human subjects. This is accomplished by asking them (via questionnaires, interviews, etc.) The methodology allows testing of a priori hypotheses and offers an iterative approach to the generation of hypotheses (Jenkins, 1985).

Participative research:
This methodology, also referred to as "Action Research", allows the researcher to become a part of the research - to be affected by and to affect the research. The objective with this methodology is not the finite testing of a particular hypothesis but the realization of the "human creative potential". Human subjects in this methodology are "of the essence" (Jenkins, 1985).

Philosophical research:
This methodology defines a purely mental pursuit. The researcher thinks and logically reasons causal relationships. The process is intellectual and the aim is for the flow of logic to be explicit, replicable and testable by others (Jenkins, 1985).

PLS: Partial Least Squares.
A second generation regression model that combines a factor analysis with linear regressions, making only minimal distribution assumptions.

PCA: Principal Components Analysis.
Statistical procedure employed to resolve a set of correlated variables into a smaller group of uncorrelated or orthogonal factors (also known as a 90 degree rotation). This is accomplished by choosing a rotation in the statistical package such as Varimax. There is no reason a researcher should ever choose an oblique rotation since we are interested in finding independent factors, and this is achieved with an orthogonal rotation. Authors (or reviewers, for that matter) who choose an oblique rotation have not theoretical reason for choosing any angle other than 90 degrees. What would be the grounds for such an argument? It is hard to imagine such a justification, but one occasionally encounters this viewpoint.

Q-sorting:
Q-sorting offers a powerful, theoretically grounded, and quantitative tool for examining opinions and attitudes. Q-sorting consists of a modified rank-ordering procedure in which stimuli are placed in an order that is significant from the standpoint of a person operating under specified conditions. It results in the captured patterns of respondents to the stimulus presented, a topic on which opinions vary. Those patterns can then be analyzed to discover groupings of response patterns, supporting effective inductive reasoning (Thomas and Watson, 2002).

Reliability:
Extent to which a variable or set of variables is consistent in what it is intended to measure. If multiple measurements are taken, the reliable measures will all be very consistent in their values.

R-squared or R²: Coefficient of determination.
Measure of the proportion of the variance of the dependent variable about its mean that is explained by the independent variable(s). R-squared is derived from the F statistic. This statistic is usually employed in linear regression analysis and PLS. In LISREL, the equivalent statistic is known as a squared multiple correlation.

Secondary data sources:
Data that was already collected for some other purpose is called secondary data. Organization files and library holdings are the most frequently used secondary sources of data. Statistical compendia, movie film, printed literature, audio tapes, and computer files are also widely used sources. Secondary data sources can be usually found quickly and cheaply. Sometimes there is no alternative to secondary sources, for example, census reports and industry statistics. Secondary data also extend the time and space range, for example, collection of past data or data about foreign countries (Emory, 1980).

SEM: Structural Equation Modeling.
Multivariate technique combining aspects of multiple regression (examining dependence relationships) and factor analysis (representing unmeasured concepts with multiple variables) to estimate a series of interrelated dependence relationships simultaneously. It is characterized by two basic components: (1) the structural model and (2) the measurement model. The structural model is the "path" model, which relates independent to dependent variables. In such situations, theory, prior experience, or other guidelines allow the researcher to distinguish which independent variables predict each dependent variable (Hair et al., 1995).

Structural model:
Set of one or more dependence relationships linking the model constructs. The structural model is most useful in representing the interrelationships of variables between dependence relationships.

Survey:
In a survey, the researcher seeks verbal or written responses to questions or statements. Surveys can be very effective in gathering data about individual preferences, expectations, past events, and private behaviors. The versatility of this method is its greatest strength. It is the only practical way to learn many types of information and the most economical way in many other situations (Emory, 1980).

Wilks' Lambda:
One of the four principal statistics for testing the null hypothesis in MANOVA. It is also referred to as the maximum likelihood criterion or U statistic (Hair et al., 1995).