Glossary

Numbers in parentheses indicate the chapter in which the term is introduced.

A-B design. A single-case design with two phases: A represents the baseline phase, and B represents the intervention phase. (12)

A-B-A design. A single-case withdrawal design in which a second baseline phase is introduced. (12)

absolute risk increase (ARI). The increase in risk associated with an intervention as compared to risk without the intervention (or control condition); the absolute difference between the control event rate (CER) and the experimental event rate (EER). (28)

absolute risk reduction (ARR). The reduction in risk associated with an intervention as compared to the risk without the intervention (or the control condition); the absolute difference between the experimental event rate (EER) and the control event rate (CER). (28)

accessible population. The actual population of subjects available to be chosen for a study. This group is usually a nonrandom subset of the target population. (8)

active variable. An independent variable with levels that can be manipulated and assigned by the researcher. (9)

adjusted means. Means that have been adjusted based on the value of a covariate in an analysis of covariance. (24)

agreement. (See percent agreement.)

alpha coefficient. (See Cronbach's alpha.)

alpha level (α). Level of statistical significance, or risk of Type I error; maximum probability level that can be achieved in a statistical test to reject the null hypothesis. Symbols α₁ and α₂ are used to denote level of significance for one- and two-tailed tests, respectively. (18)

alphanumeric data. In data processing, the entry of values that contain symbols or letters. (30)

alternate forms reliability. Reliability of two equivalent forms of a measuring instrument. (5, 26)

alternating treatment design. A single-case design in which two (or more) treatments are compared by alternating them within a session (or in alternate sessions). (12)

alternative hypothesis (H₁). Hypothesis stating the expected relationship between independent and dependent variables; considered the negation of the null hypothesis. The alternative hypothesis is accepted when the null hypothesis is rejected. (18)

analysis of covariance (ANCOVA). Statistical procedure used to compare two or more treatment groups while controlling for the effect of one or more confounding variables (called covariates). (24)

analysis of variance (ANOVA). Statistical procedure appropriate for comparison of three or more treatment groups or conditions, or the simultaneous manipulation of two or more independent variables; based on the F statistic. (20)

a priori comparisons. (See planned comparisons.)

area probability sample. A form of cluster sampling in which geographic areas serve as the units of analysis. (8)

ARIMA (autoregressive integrated moving average). Statistical technique for analysis of data from time-series studies. (11)

attribute variable. An independent variable with levels that cannot be manipulated or assigned by the researcher, but that represent subject characteristics (such as age and sex). (9)

attrition (experimental mortality). A threat to internal validity, referring to the differential loss of participants during the course of data collection, potentially introducing bias by changing the composition of the sample. (9)

audit trail. Comprehensive process of documenting interpretation of qualitative data. (14)

autocorrelation. Correlation of consecutive data points in a time series design. (12)

backward selection. A process used in stepwise multiple regression that enters all independent variables into the equation, and then removes nonsignificant variables in successive steps, until all remaining variables are significant. (29)

Bayes' theorem. The calculation of the probability of one event based on the probability of another event; used to estimate posterior (posttest) probabilities based on prior (pretest) probabilities of a diagnostic outcome. (27)

beta (β). Probability of making a Type II error. (18)

beta weight. In a multiple regression equation, the standardized weight for each independent variable. (29)

between-groups variance. That portion of the total variance in a set of scores that is attributed to the difference between groups. (19, 20)

between subjects design. An experimental design that is based on comparison between independent groups. (10)

between subjects factor. An independent variable for which levels are applied to independent groups. (20)

bimodal distribution. A distribution having two modes. (17)

binomial variable. (See dichotomy.)

bivariate statistics. Statistics involving the analysis of two variables for the purpose of determining the relationship between them, for example, correlation. (23)

blinding. Techniques to reduce experimental bias by keeping the subjects and/or investigators ignorant of group assignments and research hypotheses. (9)

block. Level of an attribute variable in which subjects are homogeneous on a particular characteristic. (9,10)

Bonferroni's adjustment (correction). A correction often used when multiple t-tests are performed, to reduce Type I error. The desired level of significance (α) is divided by the number of comparisons. The resulting value is then used as the level of signficance for each comparison to reject the null hypothesis. (21)

Bonferroni t. A post hoc method to compare means following an analysis of variance; based on planned comparisons; also called the Dunn multiple comparison procedure. (21)

Boolean logic. In literature searches, the terms AND, NOT and OR used to expand or narrow search terms. (31)

box plot. Also called box and whisker plot. A graphic display of a distribution, showing the median, 25th and 75th percentiles, and highest and lowest scores. (17)

canonical correlation. A multivariate correlation procedure, whereby two sets of variables are correlated. (29)

case-control study. A design in analytic epidemiology in which the investigator selects subjects on the basis of their having or not having a particular disease and then determines their previous exposure. (13)

ceiling effect. A measurement limitation of an instrument whereby the scale cannot determine increased performance beyond a certain level. (6)

celeration line. In single-case research, a line that divides the data points within a phase into two equal halves, indicating the trend of the data within that phase. (12)

censored observation. An observation whose value is unknown because the subject has not been in the study long enough for the outcome to have occurred; used to estimate survival curves. (29)

central tendency. Descriptive statistics that represent "averages" or scores that are representative of a distribution; includes mean, median, and mode. (17)

centroid. A point determined from the intersection of two means of two dependent variables (X, Y), used in multivariate analysis. (29)

change score. Difference between two measurements taken at different times, typically between pretest and posttest or followup. Also called a gain score. (6, 27)

chi square test (χ²). A nonparametric test applied to nominal data, comparing observed frequencies within categories to frequencies expected by chance. (25)

clinical prediction rule. A statistical tool that quantifies the relative contribution of examination and history findings to determine a diagnosis, prognosis, or likely response to intervention. (27)

cluster analysis. A multivariate statistical procedure that classifies subjects into homogeneous subsets. (29)

cluster sampling. A form of probability sampling in which large subgroups (clusters) are randomly selected first, and then smaller units from these clusters are successively chosen; also called multistage sampling. (8)

coefficient of determination (r²). Coefficient representing the amount of variance in one variable (Y) that can be explained (accounted for) by a second variable (X). (24)

coefficient of variation (CV). A measure of relative variation; based on the standard deviation divided by the mean, expressed as a percentage. (17, 26)

cohort study. An observational study design in which a specific group is followed over time. Subjects are classified according to whether they do or do not have a particular risk factor or exposure and followed to determine disease outcomes. (13)

common cause variation. Fluctuation in response resulting in random and expected variation in performance. (12)

completer analysis. Analysis of data in a clinical trial only for those subjects who complete the study. (9)

concurrent validity. A type of measurement validity; a form of criterion-related validity; the degree to which the outcomes of one test correlate with outcomes on a criterion test, when both tests are given at relatively the same time. (6)

confidence interval (Cl). The range of values within which a population parameter is estimated to fall, with a specific level of confidence. (18, 19)

confounding. The contaminating effect of extraneous variables on interpretation of the relationship between independent and dependent variables. (9, 28)

confounding variable. A variable that is more likely to be present in one group of subjects than another, and that is related to the outcome of interest, thereby potentially "confounding" interpretation of the outcome. (28)

consecutive sampling. A form of nonprobability sampling, where subjects are recruited as they become available. (8)

constant comparative method. Inductive process in qualitative research that calls for continual testing of a theory as data are examined. (14)

construct validity. 1. A type of measurement validity; the degree to which a theoretical construct is measured by an instrument. (6) 2. Design validity related to operational definitions of independent and dependent variables. (9)

content analysis. A procedure for analyzing and coding narrative data in a systematic way. (13)

content validity. A type of measurement validity; the degree to which the items in an instrument adequately reflect the content domain being measured. (6)

contingency table. A two-dimensional table displaying frequencies or counts, with rows (R) and columns (C) representing categories of nominal or ordinal variables; also referred to as cross-tabulation. (25)

continuous variable. A quantitative variable that can theoretically take on values along a continuum. (4)

control event rate (CER). The number of subjects in the control group who develop the outcome of interest. (28)

convenience sampling. A nonprobability sampling procedure, involving selection of the most available subjects for a study. (8)

convergent validity. An approach in construct validation, assessing the degree to which two different instruments or methods are able to measure the same construct. (6)

correlation. The tendency for variation in one variable to be related to variation in a second variable; those statistical procedures used to assess the degree of covariation between two variables. (23)

correlational research. A descriptive research approach that explores the relationship among variables without active manipulation of variables by the researcher. (13)

counterbalancing. Systematic alternation of the order of treatment conditions, to avoid order effects in a repeated measures design. (10)

covariate. An extraneous variable that is statistically controlled in an analysis of covariance, so that the relationship between the independent and dependent variables is analyzed with the effect of the extraneous factor removed. (24)

Cox's proportional hazards regression. A regression procedure used when the outcome has not yet occurred (a censored variable). Used in survival analysis. (29)

criterion-related validity. A type of measurement validity; the degree to which the outcomes of one test correlate with outcomes on a criterion test; can be assessed as concurrent validity or predictive validity. (6)

criterion-referencing. Interpretation of a score based on its actual value. (6)

critical value. The value of a test statistic that must be exceeded for the null hypothesis to be rejected; the value of a statistic that separates the critical region; the value that defines a statistically significant result at the set alpha level. (18)

Cronbach's alpha. Reliability index of internal consistency, on a scale of 0.00 to 1.00. (5, 26)

crossover design. A repeated measures design used to control order effects when comparing two treatments, where half of the sample receives treatment A first followed by treatment B, and the other half receives treatment B first followed by treatment A. (10)

cross-sectional study. A study based on observations of different age or developmental groups at one point in time, providing the basis for inferring trends over time. (13)

cross-tabulation. (See contingency table.)

crude rate. A rate for a population that is not adjusted for any subset of the population. (28)

cumulative incidence (Cl). The number of new cases of a disease during a specified time period divided by the total number of people at risk; the proportion of new cases of a disease in a population. (28)

cumulative scale. A scale designed so that agreement with higher-level responses assumes agreement with all lower-level responses. (15)

curvilinear relationship. The relationship between two variables that does not follow a linear proportional relationship. (23)

cut-off score. Score used as the demarcation of a positive or negative test outcome. (27)

deductive reasoning. The logical process of developing specific hypotheses based on general principles. (1)

degrees of freedom (df). Statistical concept indicating the number of values within a distribution that are free to vary, given restrictions on the data set; usually n – 1. (18) For analysis of variance, df_e = error degrees of freedom; df_b = between groups degrees of freedom; df_t = total degrees of freedom. (20)

Delphi survey. Survey method whereby decisions on items are based on consensus of a panel. (14)

dependent variable. A response variable that is assumed to depend on or be caused by another (independent) variable. (7)

developmental research. A descriptive research approach designed to document how certain groups change over time on specific variables. (14)

deviation score . The distance of a single data point from the mean of the distribution. The sum of the deviation scores for a given distribution will always equal zero. (17)

dichotomy (dichotomous variable). A nominal variable having only two categories, such as yes/no and male/female; a binomial variable. (7)

difference score (d). The difference between two scores taken on the same individual. (19)

directional hypothesis. A research hypothesis (or alternative hypothesis) that predicts the direction of a relationship between two variables. (7,18)

discrete variable. A variable that can only be measured in separate units and that cannot be measured in intervals of less than 1. (4)

discriminant analysis. A multivariate statistical technique used to determine if a set of variables can predict group membership. (29)

discriminant validity. An approach in construct validation assessing the degree to which an instrument yields different results when measuring two different constructs; that is, the ability to discriminate between the constructs. (6)

double-blind study. An experiment in which both the investigator and the subject are kept ignorant of group assignment. (9)

dummy variable (coding). In regression procedures, the assignment of codes to a nominal variable, reflecting the presence or absence of certain traits. (29)

effect size. A statistical expression of the magnitude of the difference between two treatments or the magnitude of a relationship between two variables, based on the proportional relationship of the difference to the variance. (18, 27, Appendix C)

effectiveness. Benefits of an intervention as tested under "real world" conditions, often using quasi-experimental methods. (10)

efficacy. Benefit of an intervention as tested under controlled experimental conditions, usually with a control group in a randomized controlled trial. (10)

eigenvalue. A measure of the proportion of the total variance accounted for by a factor in a factor analysis. (29)

epidemiology. Study of the distribution of disease in relation to person, place and time, and measures of risk associated with exposures to disease. (28)

error variance. That portion of the total variance in a data set that cannot be attributed to treatment effects, but that is due to differences between subjects. (19)

ethnography. An approach to qualitative research in which the experiences of a specific cultural group are studied. (14)

event rate. The proportion of subjects in a group in whom a specific event or outcome is observed. (See control event rate [CER] and experimental event rate [EER].) (28)

evidence-based practice. The application of clinical decision making for patient management based on research evidence, clinical expertise, patient values and preferences and clinical circumstances. (1)

exempt review. Exemption from review of a proposal by an Institutional Review Board for projects that do not involve direct contact with subjects, presenting no risk. (3)

expedited review. Accelerated review of a proposal by an Institutional Review Board, based on minimal risk. (3)

expected frequencies. In a contingency table, the frequencies that would be expected if the null hypothesis is true; frequencies that are expected just by chance. (25)

experimental event rate (EER). The number of subjects in the experimental or treatment group who develop the outcome of interest. (28)

experimenter effects (experimenter bias). Biases that are present in research data because of behaviors, expectations, or attitudes of those collecting the data. (9)

explained variance. Between-groups variance; that portion of the total variance in a data set that can be attributed to the differences between groups or treatment conditions. (19)

exploratory research. Research that has as its purpose the exploration of data to determine relationships among variables. (13)

external validity. The degree to which results of a study can be generalized to persons or settings outside the experimental situation. (9)

extraneous variable. A variable that confounds the relationship between the independent and dependent variables. (9)

facets. In generalizability theory, specific conditions under which reliability of a measurement can be generalized. (5, 26)

face validity. The assumption of validity of a measuring instrument based on its appearance as a reasonable measure of a given variable. (6)

factor. 1. A variable. (7) 2. A set of interrelated variables in a factor analysis. (29)

factor analysis. An exploratory multivariate statistical technique used to examine the structure within a large set of variables and to determine the underlying dimensions that exist within that set of variables. (6, 29)

factorial design. An experimental design involving two or more independent variables, allowing for the interpretation of main effects and interaction effects. (10)

false negative. A test result that is negative in a person who has the disease or condition of interest. (27)

false positive. A test result that is positive in a person who does not have the disease or condition of interest. (27)

Fisher's exact test. A nonparametric procedure applied to nominal data in a 2 × 2 contingency table, comparing observed frequencies within categories to frequencies expected by chance. Used when samples are too small to use the chi-square test. (25)

floor effect. A measurement limitation of an instrument whereby the scale cannot determine decreased performance beyond a certain level. (6)

forward selection. A process used in stepwise multiple regression that enters variables one at a time into the equation based on the strength of their association with the outcome variable, until all statistically significant variables are included. (29)

frequency distribution. A list of values that occur in a distribution, with a count of the number of times each value occurs. (17)

Friedman two-way analysis of variance by ranks (χ²_r). A nonparametric statistical procedure for repeated measures, comparing more than two treatment conditions of one independent variable; analogous to the one-way repeated measures analysis of variance. (22)

gain score. (See change score.)

Gaussian distribution. (See normal distribution.)

generalizability. 1. The quality of research that justifies inference of outcomes to groups or situations other than those directly involved in the investigation. (9) 2. The concept of reliability theory in which measurement error is viewed as multidimensional and must be interpreted under specific measurement conditions. (5, 26)

gold standard. A measurement that defines the true value of a variable. In criterion-related validity, an instrument that is considered a valid measure and that can be used as the standard for assessing validity of other instruments. (6) In diagnostic testing, a procedure that accurately identifies the true disease condition (negative or positive) of the subject. (27)

goodness of fit test. Use of chi square to determine if an observed distribution of categorical variables fits a given theoretical distribution. (25)

grand mean. The mean of all scores across groups in an analysis of variance. (20)

grounded theory. An approach to collecting and analyzing data in qualitative research, with the goal of developing theories to explain observations and experience. (14)

Guttman scale. (See cumulative scale.)

Hawthorne effect. The effect of subjects' knowledge that they are part of a study on their performance. (9)

hazard function. The probability that a subject will achieve a specific outcome in a certain time interval. (29)

histogram. A bar graph of a frequency distribution. (17)

historical controls. Subjects from previous research studies that serve as controls for experimental subjects in a subsequent study. (11)

historical research. Research that seeks to examine relationships and facts based on documentation of past events. (13)

history effect. A threat to internal validity, referring to the occurrence of extraneous events prior to a posttest that can affect the dependent variable. (9)

homogeneity of variance. An underlying assumption in parametric statistics that variances of samples are not significantly different. (18,19)

hypothesis. A statement of the expected relationship between variables. (7)

incidence. The proportion of people who develop a given disease or condition within a specified time period. (28)

independent factor. An independent variable in which the levels represent independent groups of subjects. (7)

independent variable. The variable that is presumed to cause, explain or influence a dependent variable; a variable that is manipulated or controlled by the researcher, who sets its "values" or levels. (7)

inductive reasoning. The logical process of developing generalizations based on specific observations or facts. (1)

inferential statistics. That branch of statistics concerned with testing hypotheses and using sample data to make generalizations concerning populations. (18)

informed consent. An ethical principle that requires obtaining the consent of the individual to participate in a study based on full prior disclosure of risks and benefits. (3)

instrumentation effect. A threat to internal validity in which bias is introduced by an unreliable or inaccurate measurement system. (9)

intention-to-treat. Principle whereby data are analyzed according to group assignments, regardless of how subjects actually completed the study. (9)

interaction effect. The combined effect of two or more independent variables on a dependent variable. (10, 20)

intercorrelations. A set of bivariate correlations for several variables within a sample. (23)

internal consistency. A form of reliability, assessing the degree to which a set of items in an instrument all measure the same trait. Typically measured using Cronbach's alpha. (5, 26)

internal validity. The degree to which the relationship between the independent and dependent variables is free from the effects of extraneous factors. (9)

interquartile range. The difference between the first and third quartiles in a distribution, often expressed graphically in a boxplot. (17)

interrater reliability. The degree to which two or more raters can obtain the same ratings for a given variable. (5, 26)

interrupted time-series design. A design involving a series of measurements over time, interrupted by one or more treatment occasions. (10)

interval scale. Level of measurement in which values have equal intervals, but no true zero point. (4)

intraclass correlation coefficient (ICC). A reliability coefficient based on an analysis of variance; a generalizability coefficient. (26)

intrarater reliability. The degree to which one rater can obtain the same rating on multiple occasions of measuring the same variable. (5, 26)

item-to-total correlation. Correlation of individual items in a scale with the total scale score; an indication of internal consistency. (5, 26)

Kaplan-Meier Estimate. A common method of determining survival time which generates a step function, changing the survival estimate each time a patient dies (or reaches the terminal event). (29)

kappa (κ). A correction factor for percent agreement measures of reliability, accounting for the potential effect of chance agreements. (26)

known groups method. A technique for construct validation, in which validity is determined by the degree to which an instrument can demonstrate different scores for groups known to vary on the variable being measured. (6)

Kruskal-Wallis one-way analysis of variance by ranks (H). A nonparametric statistical procedure for comparing more than two independent groups representing levels of one independent variable; analogous to the one-way analysis of variance. (22)

Latin square. A matrix of columns and rows used to assign sequences of treatments to control for order effects. (10)

Least squares method. A method of fitting a regression line to a set of bivariate data so as to minimize the sum of the squared vertical deviations of Y values around that line. (24)

level. 1. The "value" or classification of an independent variable. (7) 2. In single-case research, the magnitude of the target behavior; changes in level are associated with differences in magnitude between the end of one phase and the beginning of the following phase. (12)

level of measurement. The precision of a scale based on how a characteristic is measured; nominal, ordinal, interval and ratio levels. (4)

level of significance (α). The probability that an observed effect could be attributed to chance; the standard for rejecting the null hypothesis; traditionally set at a = .05. (18)

Levene's test. A test of the equality of variances, used with the independent t test and the analysis of variance. (19)

Likert scale. A summative scale based on responses to a set of statements for which respondents are asked to rate their degree of agreement or disagreement. (15)

likelihood ratio. In diagnostic testing, the ratio indicating the usefulness of the test for ruling in or ruling out a condition. (See negative likelihood ratio and positive likelihood ratio.) (27)

limits of agreement. Index of reliability between alternate forms of an instrument. (26)

linear regression. The process of determining a regression equation to predict values of Y based on a linear relationship with values of X. (24)

line of best fit. The regression line, representing the relationship between two variables, usually plotted on a scatter diagram. (24)

logistic regression. Multiple regression procedure where the dependent variable is a dichotomous outcome; predicts odds associated with presence or absence of the dependent variable based on the independent variables. (29)

logrank test. A statistical procedure for comparing two survival curves when censored observations are present. (29)

longitudinal study. A study designed to collect data over time, usually for the purpose of describing developmental changes in a particular group. (13)

main effect. The separate effect of one independent variable in a multifactor design. (10, 20)

Mann-Whitney U test. A nonparametric statistical test for comparing two independent groups; analogous to the unpaired t-test. (22)

MANOVA. (See multivariate analysis of variance.)

maturation effect. A threat to internal validity, in which changes occur in the dependent variable as a result of the passing of time. (9)

McNemar test. A nonparametric statistical test for nominal level measures, for correlated samples; a form of the chi square test. (25)

mean (X̄). A measure of central tendency, computed by summing the values of several observations and dividing by the number of observations. (17)

mean square (MS). In an analysis of variance, that value representing the variance; calculated by dividing the sum of squares for a particular effect by the degrees of freedom for that effect. The symbol MS_e = error mean square; MS_b, = between groups mean square. (20)

measurement error. The difference between an observed value for a measurement and the theoretical true score; may be the result of systematic or random effects. (5, 26)

median. A measure of central tendency representing the 50th percentile in a ranked distribution of scores; that is, that point at which 50 percent of the scores fall below and 50 percent fall above. (17)

meta-analysis. Use of statistical techniques in a systematic review to integrate the results of included studies to determine overall outcome, usually based on effect size. (16)

method error (ME). A form of reliability testing for assessing response stability based on the discrepancy between two sets of repeated scores. (26)

methodological research. Research designed to develop or refine procedures or instruments for measuring variables, generally focusing on reliability and validity. (13)

minimal clinically important difference (MCID). The smallest difference in a measured variable that signifies an important rather than trivial difference in the patient's condition. The smallest difference a patient or clinician would perceive as beneficial, and that would result in a change in the management of the patient. Also called minimal clinically important change (MCIC) or minimally important change (MIC). (6, 27)

minimal detectable difference (MDD). That amount of change in a variable that must be achieved to reflect a true difference; the smallest amount of change that passes the threshold of error. Also called minimal detectable change (MDC). (5, 26)

mixed design. A design that incorporates independent variables that are independent (between-subjects) and repeated (within-subjects) factors. Also called a split-plot design. (10)

mode. A measure of central tendency representing the most commonly occurring score. (17)

μ (mu). Mean of a population. (17)

multicolinearity. The correlation between independent variables in a multiple regression equation, causing them to provide redundant information. (29)

multiple baseline design. In single-case research, a design for collecting data for more than one subject, behavior, or treatment condition. Baseline phases are staggered to provide control. (12)

multiple comparison test. A test of differences between individual means following analysis of variance, used to control for Type I error. (21)

multiple regression. A multivariate statistical technique for establishing the predictive relationship between one dependent variable (Y) and a set of independent variables (X₁), X₂,…). (29)

multistage sampling. (See cluster sampling.)

multivariate analysis. A set of statistical procedures designed to analyze the relationship among three or more variables; includes techniques such as multiple regression, discriminant analysis, factor analysis and multivariate analysis of variance. (29)

multivariate analysis of variance (MANOVA). An advanced multivariate procedure that provides a global test of significance for multiple dependent variables using an analysis of variance. (29)

natural history. Longitudinal study of a disease or disorder, demonstrating the typical progress of the condition. (14)

naturalistic inquiry. Qualitative observation and interaction with subjects in their own natural environment. (14)

negative likelihood ratio (LR –). A ratio that indicates how much the odds of a disease are decreased if a diagnostic test is negative. Equals specificity/1 – sensitivity. (27)

negative predictive value (PV –). In diagnostic testing, the proportion of subjects who are correctly identified as not having the condition of interest. (27)

nested design. A multifactor design in which one variable is not crossed with other variables. (10)

Newman-Keuls (NK) multiple comparison test. A multiple comparison procedure, used following a significant analysis of variance. Also called the Student-Newman-Keuls (SNK) test. (21)

nominal scale. Level of measurement for classification variables; assignment of "values" based on mutually exclusive and exhaustive categories with no inherent rank order. (4)

nondirectional hypothesis. A research hypothesis (or alternative hypothesis) that does not indicate the expected direction of the relationship between variables. (7,18)

nonequivalent control group. A control group (or comparison group) that was not created by random assignment. (11)

nonparametric statistics. A set of statistical procedures that are not based on assumptions about population parameters, or the shape of the underlying population distribution; most often used when data are measured on the nominal or ordinal scales. (22)

nonprobability sample. A sample that was not selected using random selection. (8)

normal distribution (curve). A symmetrical bell-shaped theoretical distribution that has defined properties; also called a Gaussian distribution. (17,18)

normative research. A descriptive research approach designed to determine normal values for specific variables within a population. (14)

norm referencing. Interpretation of a score based on its value relative to a standard or "normal" score. (6)

null hypothesis (H₀). A statement of no difference or no relationship between variables; the statistical hypothesis. (7, 18)

number needed to harm (NNH). The number of patients that need to be treated to observe one adverse outcome. (28)

number needed to treat (NNT). The number of patients that need to be treated to prevent one adverse outcome or achieve one successful outcome; the reciprocal of absolute risk reduction (ARR). (28)

observational study. A study that does not involve an intervention or manipulation of an independent variable. (13)

odds ratio (OR). Estimate of relative risk in a case-control study. (28)

one-tailed test. A statistical test based on a directional alternative hypothesis, in which critical values are obtained for only one tail of a distribution. (18)

one-way analysis of variance. An analysis of variance with one independent variable. (20)

one-way design. An experimental or quasi-experimental design that involves one independent variable. (10)

on-protocol analysis. Analysis of data in an experiment based only on subjects who completed the study according to assigned groups. Also called completer analysis or on-treatment analysis. (9)

open-ended question. A question on a survey (interview or questionnaire) that does not restrict the respondent to specific choices, but allows for a free response. (15)

operational definition. Definition of a variable based on how it will be used in a particular study; how a dependent variable will be measured, how an independent variable will be manipulated. (7)

order effects. The sequential effect of one subject being exposed to several treatments in the same order; potentially manifested as carryover or practice effects. (10)

ordinal scale. Level of measurement in which scores are ranks. (4)

outlier. Numeric value that does not fall within the range of most scores in a distribution. (24)

paired t-test. A parametric test for comparing two means for correlated samples or repeated measures; also called a correlated t-test. (19)

paradigm. A set of assumptions, concepts or values that constitute a way of viewing reality within an intellectual community. (1)

parameter. A measured characteristic of a population. (17)

parametric statistics. Statistical procedures for estimating population parameters and for testing hypotheses based on population parameters, with assumptions about the distribution of variables, and for use with interval or ratio measures. (18)

partial correlation. A statistical technique for establishing the correlation between two variables, with the effect of a third variable removed; also called a first-order correlation. (29)

participant observation. A method of data collection in qualitative research in which the researcher becomes a participant in the group that is being observed. (14)

Pearson product-moment coefficient of correlation (r). A parametric statistical technique for determining the relationship between two variables. (23)

percent agreement. A reliability test for categorical variables, estimating the ability of researchers to agree on category ratings. (26)

percentile. The percentage of a distribution that is below a specified value. Data are divided into 99 equal ranks, or percentiles, with 1 percent of the scores in each rank. (17)

person-years. The total number of years that a set of subjects have participated in a study, typically used when subjects begin and end their participation at different times. (28)

phenomenology. An approach to qualitative research involving the study of complex human experience as it is actually lived. (14)

phi coefficient (r_Φ). A nonparametric correlation statistic for estimating the relationship between two dichotomous variables. (23,25)

point biserial correlation (r_pb). A correlation statistic for estimating the relationship between a dichotomy and a continuous variable on the interval or ratio scale. (23)

point estimate. A single sample statistic that serves as an estimate of a population parameter. (18)

polynomial regression. Regression procedure for nonlinear data. (24)

pooled variance estimate (s_p²). Estimate of population variance based on the weighted average of sample variances; used in the unpaired t-test when group variances are not significantly different (under conditions of homogeneity of variance). (19)

population. The entire set of individuals or units to which data will be generalized. (8)

positive likelihood ratio (LR +). A ratio that indicates how much the odds of a disease are increased if a diagnostic test is positive. Equals sensitivity/1 – specificity. (27)

positive predictive value (PV +). Estimate of the likelihood that a person who tests positive actually has the disease. (27)

posterior probability. (See posttest probability.)

post hoc comparisons. Multiple comparison tests that follow an analysis of variance. (21)

posttest-only design. An experimental design in which only one measurement is taken following treatment. (10)

posttest probability. The probability of a condition existing after performing a diagnostic test; predictive value of a diagnostic test. Also called posterior probability. Depends on the pretest probability, and the test's sensitivity and specificity. (27)

power (1 – β). The ability of a statistical test to find a significant difference that really does exist; the probability that a test will lead to rejection of the null hypothesis. (18, Appendix C)

predictive validity. A form of measurement validity in which an instrument is used to predict some future performance. (6)

predictive value (PV). (see negative predictive value and positive predictive value)

preference. In sequential clinical trials, the expression of which treatment is considered better within a sequential pair. (10)

pretest probability. The probability that a condition exists prior to performing a diagnostic test. Equal to the prevalence of the condition in a specified group of subjects. Also called prior probability. (27)

pretest-posttest design. An experimental design involving a pretest prior to intervention and a posttest following intervention. (10)

prevalence. The number of existing cases of a disease or condition at a given point in time, expressed as a proportion of the total population at risk. (27, 28)

primary source. Reference source that represents the original document by the original author. (7)

prior probability. (See pretest probability.)

probability sample. A sample chosen using randomized methods. (8)

proportional hazards model. (See Cox's regression.)

prospective study. A study designed to collect data following development of the research question. (13)

publication bias. Tendency for researchers and editors to treat positive experimental results (finding an effect) differently from negative or inconclusive results (finding no effect), often with a preference for publication of positive findings. (16)

purposive sample. A nonprobability sample in which subjects are specifically selected by the researcher on the basis of subjective judgment that they will be the most representative. (8)

q. Studentized range statistic, used in multiple comparison tests. (21)

Q-sort. An analytic technique used to characterize attitudes, opinions, or judgments of individuals through a process of comparative rank ordering. (15)

quadratic trend. A nonlinear trend, with one turn in direction. (21, 24)

quartile (Q). Three quartiles divide a distribution of ranked data into four equal groups, each containing 25 percent of the scores. (17)

quasi-experimental research. Comparative research approach in which subjects cannot be randomly assigned to groups or control groups are not used. (11)

quota sampling. Nonprobability sampling method in which stratification is used to obtain representative proportions of specific subgroups. (8)

R² (R squared). Multiple correlation coefficient squared; represents the proportion of variance in Y explained by several independent variables in a multiple regression equation. (29)

random assignment. Assignment of subjects to groups using probability methods, where every subject has an equal chance of being assigned to each group. (9)

random sampling. Probability method of selecting subjects for a sample, where every subject in the population has an equal chance of being chosen. (8)

random selection. (See random sampling.)

randomized block design. An experimental design in which one independent variable is an attribute variable, creating homogeneous blocks of subjects who are then randomly assigned to levels of the other independent variable. (10)

randomized controlled trial (RCT). An experimental study in which a clinical treatment is compared with a control condition, where subjects are randomly assigned to groups. Also called a randomized clinical trial. (10)

rank sum test. A nonparametric statistical procedure, used to compare two independent samples; equivalent to the Mann-Whitney U test. Analogous to the unpaired t-test. (22)

Rasch analysis. Transformation of items on an ordinal scale to an interval scale, demonstrating the unidimensional nature of a scale. (15)

ratio scale. The highest level of measurement, in which there are equal intervals between score units and a true zero point. (4)

reactive measurement. A measurement that distorts the variable being measured, either by the subject's awareness of being measured or by influence of the measurement process. (9)

recall bias. The possible inaccuracy of recalling medical history or previous exposures; of particular concern in retrospective studies. (28)

receiver operating characteristic (ROC) curve. In diagnostic testing, a plot of the true positives (sensitivity) against false positives (1 – specificity) at several cutoff points for defining a positive test. (27)

refereed journal. Journals that utilize a peer review process to evaluate manuscript submissions as a basis for choosing which ones will be published. (33)

reference standard. A value used as a standard against which to judge a criterion; may or may not be a gold standard. Used to judge criterion-related validity or diagnostic accuracy. (6, 27)

regression analysis. A statistical procedure for examining the predictive relationship between a dependent (criterion) variable and an independent (predictor) variable. (24, 29)

regression coefficient. In a regression equation, the weight (b) assigned to the independent variable; the slope of the regression line. (24, 29)

regression line. The straight line that is drawn on a scatter plot for bivariate data from the regression equation, summarizing the relationship between variables. (24)

regression toward the mean. A statistical phenomenon in which scores on a pretest are likely to move toward the group mean on a posttest because of inherent positive or negative measurement error; also called statistical regression. (5, 9)

relative risk (RR). Estimate of the magnitude of the association between an exposure and disease, indicating the likelihood that the exposed group will develop the disease relative to those who are not exposed. (28)

relative risk reduction (RRR). The reduction in risk associated with an intervention relative to the risk without the intervention (control); the absolute difference between the experimental event rate and the control event rate divided by the control event rate. (28)

reliability. The degree of consistency with which an instrument or rater measures a variable. (5, 26)

repeated measure (repeated factor). An independent variable for which subjects act as their own control; that is, all subjects are exposed to all levels of the variable. Also called a within- subjects factor. (7, 10)

research hypothesis. A statement of the researcher's expectations about the relationship between variables under study. (7)

residual (Y − Ŷ). In regression analysis, the difference between the value of the dependent variable predicted by the regression equation and the actual value. (24)

responsiveness. The ability of a test to demonstrate change. (6, 27)

retrospective study. A study that analyzes observations that were collected in the past. (13)

ρ (rho). Correlation coefficient for a population. (23)

risk-benefit ratio. An ethical principle that is an element of informed consent, in which the risks of a research study to the participant are evaluated in relation to the potential benefits of the study's outcomes. (3)

risk factor. A characteristic or exposure that potentially increases the likelihood of having a disease or condition. (28)

ROC curve. (See receiver operating characteristic curve.)

sampling bias. Bias that occurs when individuals who are selected for a sample overrepresent or underrepresent the underlying population characteristics. (8)

sampling distribution. A theoretical frequency distribution of a statistic, based on the value of the statistic over an infinite number of samples. (18)

sampling error. The difference between an observed statistic from a sample and the population parameter. (18)

scale of measurement. (See level of measurement.)

scatter plot. A graphic representation of the relationship between two variables. (23)

Scheffé's multiple comparison test. A multiple comparison procedure for comparing means following a significant analysis of variance. Considered the most conservative of the multiple comparison methods. (21)

secondary analysis. An approach to research involving the use of data that were collected for another purpose, usually for the purpose of testing new hypotheses. (13)

secondary source. Reference source that represents a review or report of another's work. (7)

selection bias. A threat to internal validity in which bias is introduced by initial differences between groups, when these differences are not random. (9)

semantic differential. A technique used to measure attitudes by asking respondents to rate concepts on a 7-point scale which represents a continuum across two extremes. (15)

sensitivity. A measure of validity of a screening procedure, based on the probability that someone with a disease will test positive. (27)

sensitivity analysis. A procedure in decision making to determine how decisions change as values are systematically varied. (16)

sequential clinical trial. Experimental research design that allows consecutive entrance to a clinical trial and continuous analysis of data, permitting stopping of the trial when data are sufficient to show a significant effect. (10)

serial dependency. Correlation in a set of data collected over time, in which one observation can be predicted based on previous observations. (12)

σ (sigma). Standard deviation of a population, σ² is the population variance. (17)

Σ (sigma, uppercase). Read as: "the sum of." (17)

sign test. A nonparametric statistical procedure for comparing two correlated samples, based on comparison of positive or negative outcomes; analogous to the paired t-test. (22)

significance level (α). (See alpha level.)

single-blind study. An experiment in which either the investigator or the subject is kept ignorant of group assignment, but not both. (9)

single-factor design. An experimental design involving one independent variable. (10)

single-subject design. An experimental design based on time-series data from one or more subjects, with data compared across baseline and intervention phases. Also called single-case designs. (12)

skewed distribution. A distribution of scores that is asymmetrical, with more scores to one extreme. (17)

slope. 1. In regression analysis, the rate of change in values of Y for one unit of change in X. (24) 2. In single-case research, the rate of change in the magnitude of the target behavior over time. (12)

SnNout. When a test has high sensitivity, a negative test rules out the diagnosis. (27)

snowball sampling. A nonprobability sampling method in which subjects are successively recruited by referrals from other subjects. (8)

Spearman-Brown formula. The statistical procedure used to analyze split-half reliability; also called the Spearman-Brown prophecy formula. (5)

Spearman's rank correlation coefficient (r_s). A nonparametric correlation procedure for ordinal data. Also called Spearman's rho. (23)

special cause variation. Fluctuation in response caused by known factors that results in non- random and unexpected performance.

specificity. A measure of validity of a screening procedure, based on the probability that someone who does not have a disease will test negative. (27)

split-half reliability. A reliability measure of internal consistency based on dividing the items on an instrument into two halves and correlating the results. (5)

split middle line. In single-case research, a line used to separate data points within one phase into two equal halves, reflecting the trend of the data within that phase. (See celeration line.) (12)

split-plot design. (See mixed design.)

SpPin. When a test has high specificity, a positive test rules in the diagnosis. (27)

standard deviation (s). A descriptive statistic reflecting the variability or dispersion of scores around the mean. (17)

standard error of measurement (SEM). A reliability measure of response stability, estimating the standard error in a set of repeated scores. (26)

standard error of the estimate (SEE). In regression analysis, an estimate of prediction accuracy; a measure of the spread of scores around the regression line. (24)

standard error of the mean (s_X̄). The standard deviation of a distribution of sample means; an estimate of the population standard deviation. (18)

standardized residual. In a chi square test, the contribution of each cell to the overall statistic. (25)

standardized response mean (SRM). One approach to evaluating effect size with change scores. Calculated as the difference between pretest and posttest scores, divided by the standard deviation of the change scores. (27)

standardized score. (See z-score.)

statistic. A measured characteristic of a sample. (12)

statistical conclusion validity. The validity of conclusions drawn from statistical analyses, based on the proper application of statistical tests and principles. (9)

statistical hypothesis. (See null hypothesis.)

statistical process control (SPC). A method of charting production outcomes over time to identify and monitor variances; can be used as a method of analysis for single-subject designs. (12)

statistical regression. (See regression toivard the mean.)

statistical significance. The term indicating that the results of an analysis are unlikely to be the result of chance at a specified probability level; rejection of the null hypothesis. (18)

stem-and-leaf plot. A graphic display for numerical data in a frequency distribution showing each value in the distribution. (17)

stepwise regression. An approach to multiple regression that involves a sequential process of selecting variables for inclusion in the prediction equation. (29)

stopping rule. In a sequential clinical trial, the threshold for stopping a study based on crossing a boundary that indicates a difference or no difference between treatments. (10)

stratification. The grouping of individuals in a population into homogeneous groups on some characteristic prior to sampling. (8)

sum of squares (SS). A measure of variability in a set of data, equal to the sum of squared deviation scores for a distribution ; the numerator in the formula for variance. (17). Used in analysis of variance as the basis for partitioning between-groups (SS_b) and within-groups error (SS_e) variance components. (20)

survival analysis. Multivariate analysis to estimate survival time, or time to a defined outcome, based on probabilities that an individual will achieve the outcome. (29)

systematic error. A form of measurement error, where error is constant across trials. (5)

systematic review. Review of a clearly formulated question that uses systematic and explicit methods to identify, select and critically appraise relevant research. (16)

systematic sample. A probability sampling method where subjects are chosen from lists of population members using specified intervals, such as every 10th person. (8)

t-test. A parametric test for comparing two means; also called Student's t-test. (See paired t-test and unpaired t-test.) (19)

target population. The larger population to which results of a study will be generalized. (8)

testing effect. The effect that occurs when a test itself is responsible for observed changes in the measured variable. (5, 9)

test-retest reliability. The degree to which an instrument is stable, based on repeated administrations of the test to the same individuals over a specified time interval. (5)

time-series design. A quasi-experimental design in which performance changes are assessed over time, prior to and following the administration of treatment. (11)

transformation. Mathematical conversion of a distribution to a different scale by a constant (such as square root or log) to change the shape or variance characteristics of the distribution. (Appendix D)

translational research. Clinical investigation with human subjects in which knowledge obtained from basic research is translated into diagnostic or therapeutic interventions that can be applied to treatment or prevention. (1)

treatment arm. Another term for each independent group in a clinical trial. (10)

treatment threshold. In clinical decision making, the point at which a decision is reached to treat the patient without first performing a diagnostic test. (27)

treatment-received analysis. Analysis of subject data in an experiment according to the treatment subjects actually did receive, regardless of their original group assignment. (9)

trend. 1. The shape of a distribution of scores taken over time, reflecting the distribution's linearity or lack of linearity. (21) 2. In single-case research, the direction of change in the target behavior within a phase or across phases. (12)

trend analysis. Part of an analysis of variance, used to assess trend within data taken over ordered intervals; can express data as linear, quadratic, cubic, and so on, reflecting the number of changes in direction in the data over time. (21)

triangulation. The use of multiple methods to document phenomena. (14)

true negative. A test result that is negative for those who do not have the disease or condition of interest. (27)

true positive. A test result that is positive for those who do have the disease or condition of interest. (27)

Tukey's honestly significant difference (HSD). A multiple comparison test for comparing multiple means following a significant analysis of variance. (21)

Two standard deviation band method. A method of data analysis in single-subject research.(12)

two-tailed test. A statistical test based on a nondirectional alternative hypothesis, in which critical values represent both positive and negative tails of a distribution. (18)

two-way analysis of variance. An analysis of variance with two independent variables. (20)

two-way design. An experimental or quasi-experimental study that involves two independent variables. (10)

Type I error. An incorrect decision to reject the null hypothesis, concluding that a relationship exists when in fact it does not. (18)

Type II error. An incorrect decision to accept the null hypothesis, concluding that no relationship exists when in fact it does. (18)

univariate analysis. Statistical procedures for analyzing one dependent variable. (29)

unpaired t-test. A parametric test for comparing two means for independent samples; also called an independent t-test. (19)

validity. 1. The degree to which an instrument measures what it is intended to measure. (6) 2. The degree to which a research design allows for reasonable interpretations from the data, based on controls (internal validity), appropriate definitions (construct validity), appropriate analysis procedures (statistical conclusion validity), and generalizability (external validity). (9)

variable. A characteristic that can be manipulated or observed and that can take on different values, either quantitatively or qualitatively. (2, 7)

variance (s²). A measure of variablity in a distribution, equal to the square of the standard deviation. (17)

vital statistics. Mortality and morbidity rates. (28)

washout period. In a crossover design, that period of time between administration of the two treatments, allowing effects of the experimental treatment to dissipate. (10)

weighted kappa (κ_w). An estimate of percentage agreement, corrected for chance, based on weights reflecting levels of seriousness of disagreements. (26)

Wilcoxon rank-sum test. (See Mann Whitney U test.)

Wilcoxon signed-ranks test (T). A nonparametric statistical procedure, comparing two correlated samples (repeated measures); analogous to the paired t-test. (22)

withdrawal design. In single-case research, a design that involves withdrawal of the intervention. (12)

within-groups variance. (See error variance.)

within-subjects design. A research design that incorporates only repeated measures. (10)

within-subjects factor. (See repeated measure.)

X̄. Mean of a sample. (17)

Yates' correction for continuity. In the chi square test, a correction factor applied when expected frequencies are too small, effectively reducing the chi square statistic. (25)

z distribution. The standardized normal distribution, with a mean of 0 and a standard deviation of 1. (17)

z-score The number of standard deviations that a given value is above or below the mean of the distribution; also called a standardized score. (17)

zero-order correlation. A bivariate correlation. (29)