Chapter 4 Flashcards
Observed Score
= True Score + Systematic Error + Random Error
Coefficient alpha
estimate is derived from the correlations of each item with each other item and so does not rest on any arbitrary choice of ways to divide the items into two halves. Preferred measure of internal consistency reliability
Constructs
the abstractions that social scientists discuss in their theories. They are the rich theoretical concepts that make the science interesting, terms such as social status, power, intelligence, and gender roles. Because we cannot literally put a finger on any of these concepts to measure them, we must find some concrete representations that approximate what we mean when we speak of such concepts.
Convergent Validity
Overlap between alternative measures that are intended to tap the same construct but that have different sources of systematic error.
Convergent Validity Coefficients
Correlations between scores that reflect the same trait measured by different methods.
Correlation Coefficient
a statistical index of the strength of association between 2 variables.
Definitional Operationism
the assumption that the operational definition is the construct. (Ex: intelligence is what an intelligence test measures).
Discriminant Validity
A validity measure has to show good convergence with other measures of the same thing. It should also fail to correlate with measures that are supposed to tap basically different constructs. (That’s basically the definition of discriminant validity).
Discriminant Validity Coefficient
Indicates the correlation between different traits measured by the same method.
Face Validity
evaluated by a group of judges, sometimes experts, who read or look at a measuring technique and decide whether in their opinion it measures what its name suggests.
Internal consistency reliability
alternative estimate is not subject to these concerns and, therefore, more widely used
Kappa
measure of agreement that can be used to estimate inter-rater reliability
Methods
mode of measurement.
Multitrait-multimethod matrix
table of correlation coefficients that enables us to simultaneously evaluate the convergent and discriminant validity of a construct
Nomological Net
the theoretical network of construct-to-construct associations derived from relevant theory and stated at an abstract level (Look at pg. 78 – 79 in book)
Nonsense Coefficient
Reflects the correlation between different traits measured by different methods.
Operational Definition
specifies how to measure a variable so that we can assign someone a score such as high, medium, or low social power.
Random Error
reflects nonsystematic, ever-changing influences on the score
Reliability
the extent to which a measure is free from random error.
Reliability Coefficients
correlations between scores that reflect the same trait and the same method. Although these coefficients are not themselves indicative of validity, they indicate the limit to the validity of our measure.
Social Desirability Response Bias
general tendency to over report one’s desirable behaviors and other characteristics and to under report one’s less admirable qualities.
Split-half reliability
set of items in the measure is split into two halves, and the split is usually done by separating the full set of items into odd-numbered and even-numbered sets; a strategy that ensures an equivalent number of items from early and late in the measure appear in the two sets.
Systematic Error
reflects influences from other constructs besides the desired one
Test – ReTest Correlation
provides an estimate of the measure’s reliability. (Ex: Correlation of several tests averaged together).
Traits
the underlying construct the measurement is supposed to tap (Ex: attitudes towards women).
True Score
function of the construct we are attempting to measure
Validity
the extent to which a measure reflects only the desired construct without contamination from other systematically varying constructs. (Note: Validity requires reliability as a pre-requisite).
Variables
representations of constructs. They cannot be synonymous with a construct because any single construct has many different variables. Therefore, variables are partial, fallible representations of constructs, and we work with them because they are measurable. They suggest ways in which we can decide whether someone has more or less of the construct.
