Glossary of Terms

a statistical test which compares the distribution of two or more sample groups to determine if one or more of the groups are significantly different from the others

Computer Aided Telephone Interviewing. Using a computer and specialized software to systematically interview respondents and record their answers.

Participant confidentiality exists when the researcher knows information about the respondent's identity but does not share that information. Anonymity is where all identifying information about the participant is unknown to the researcher.

outcome of an attempt to reach someone selected to complete a survey (completed, refused, answering machine, unable, etc.)

An analytical procedure that can be used for identifying the number and nature of constructs underlying a set of measures

a standardized tool used by a researcher to gather information from respondents. This may take the form of a closed-ended questionnaire, a list of open-ended questions, a personality inventory, etc.

Branching and Skip Logic allow participants that respond differently to certain questions to be routed to another sequence of questions, often through programming in an electronic (web-based, CATI) questionnaire. It can also be used to randomize question order or ask only those questions applicable to the respondent.

A variant of multiple regression, used when the dependent variable is a dichotomy, such as success/failure

describes how the research will be done, which makes research reproducible

An equation or analysis where two or more independent (predictor) variables affect the dependent variable. It tells a researcher about the relative impact that each predictor variable has on the relationship between an independent and a dependent variable.

any individual member of a sample selected to participate in study

analytic approach that uses iterative estimations to more accurately calculate scores. Transforms ordinal data into interval data so that statistical processes do not violate mathematical assumptions.

Random Sampling - A sampling method in which all elements in the population have an equal chance of being selected. Random samples have important properties that are necessary in many statistical tests.
Stratified Sampling - participants are selected at random from mutually exclusive subgroups, for example drawing a random sample based on a personal characteristic such as age group, ethnic group or sex. Stratified Sampling helps ensure that subgroups are adequately represented in the study.

The degree to which an instrument will yield the same result when applied to the same participant or sample of participants more than once. Equally, reliability is achieved when an instrument can be used in different settings or populations so long as each administration does not differ on any relevant variables.

the percentage of qualified participants in a survey sample that take part in the study.

segment of the population selected to participate in a study (see "Sampling" for more information).

When investigating social phenomena, the most accurate answer to any question would be found by questioning every member of the population. Surveying an entire population is usually too time-consuming or costly, but inferences can be made about the population by surveying a sample of the population. Sampling methods can be divided into two major categories: "probabilistic" and "non-probabilistic"

the gap between statistics reported from the sample and the actual statistic in the population. A larger sample or higher response rate within a study will reduce sampling error and allow a researcher to be more confident that data collected from a sample better reflects the population.

Chi-square - a measure of confidence that a researcher can have in using one variable to predict another. Determine if the patterns exhibited by data could have been produced by chance
T-Test - A test for comparing the mean values from two samples (different groups on same test or same group at different times). It shows how confident we can be that the two mean values differ

the likelihood that an observed relationship among variables is actually present rather than a fluke of sampling or measurement. Researchers use significance tests to calculate the chance that we observe a relationship that does not actually exist - a "Type I error." Significance is usually referred to in terms of "confidence" or "confidence levels." In media polling, it is often represented as "plus or minus n%."

The degree to which an instrument measures what it claims to measure, for example satisfaction with services or attitudes towards an issue. Note: A measure can be reliable without being valid, but cannot be valid without being reliable!

Dependent variable - A symbol or concept expected to be explained or caused by the independent variable. Also known as criterion variable.
Independent variable - A variable that is thought to influence the dependent variable.

Variables fall into one of four categories:

Nominal - responses do not have any rank order, but rather are names for categories. For example, gender or region of the country
Ordinal - responses can be rank-ordered, but the distance between each category is not consistent. "Good, better, best" as response categories can be ranked, but the distance between "good" and "better" is not necessarily the same distance as between "better" and "best."
Interval - responses can be rank-ordered, and the distance between any two consecutive responses is the same as any two other consecutive responses. Zero does not necessarily mean that absence of the variable (e.g., temperature of 0 degrees)
Ratio - continuous data where both differences between values and ratios are interpretable. Ratio data has a natural zero. A good example is birth weight in kg.