Research Psychology, Exam 2 Notes
Research Psychology, Exam 2 Notes 1094
Popular in Principles Psychology Research
Popular in Psychology
This 7 page Class Notes was uploaded by Logan Mehalic on Wednesday September 7, 2016. The Class Notes belongs to 1094 at Virginia Polytechnic Institute and State University taught by Robert J. Harvey in Fall 2016. Since its upload, it has received 6 views. For similar materials see Principles Psychology Research in Psychology at Virginia Polytechnic Institute and State University.
Reviews for Research Psychology, Exam 2 Notes
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 09/07/16
Chapter 4 Data and the Nature of Measurement -Constructs v. Measured variables Our objective it to obtain scores on one or more psychological concepts Trying to measure things you cant directly observe and things that are hypothetical Indirect approach to measurement we measure things we can observe (behaviors) and work backwards If we wanted to measure the construct of intelligence you give a test and work backwards Constructs cause the item responses Constructs aren’t real, but we assume they are causal There is some degree of error causing the item responses -Research variables Psychological research is about examining relationships between variables Variable: any characteristic that can take on more than one value o Examples: speed, level of hostility, accuracy of feedback, reaction time Research is the study of the relationship between variables o Therefore, there must be at least two variables in a research study (or there is no relationship to study) The key thing about variables is that they vary, different people get different scores on them To examine relationship they have to vary, and there needs to be at least two of them -Measuring variables in research Measurement: a process by which we assign numbers to indicate the amount of some variable present Sometimes the number assignment is easy to understand (e.g. speed measured in number of seconds) Sometimes it is more arbitrary (e.g. 1 for male and 2 for female) Our metrics are usually arbitrary in research psych -Basic scaling issues What sort of information is provided by the numbers we assign to objects? Information conveyed by scales can be grouped into two main categories o Differences between objects in “kind” o Differences in degree/amount (tests, personality, ability, achievement) how much of the characteristic someone has Level of information is numbers determines the kinds of inferences we can draw The first question we need to ask is what information you are trying to get out of it Matters because the level of information we assume a scale produces determines the kinds of inferences we can draw about people based on the scores that we obtain on that scale -Stevens’ scales Taxonomy for levels of measurements (one of the most popular) 4 types of psychological scales o Nominal o Ordinal o Interval o Ratio Second column is property of objects being conveyed Third column is permissible operations Fourth column is permissible data transformations Horizontal lines o Red, above is categorical, below is quantitative measurement As you move down the permissible data transformations, they become more and more restrictive as we move down the page and get better levels of measurements (subsets that involve additional constraints) Permissible operations work the other way, moving down the page there are more things we can do to the data the better the level of measurement gets o Nominal scales conveys the least information Ordinal scales give relative rank/order o Ex. Anything in a rank, race, high school class, percentiles Now we know category and rank (greater than or less than) Interval and Ratio are desired in research Each higher level of measurement we have allows us to build on the previous measurement Interval allows category, rank, and size of difference between the scales units (basic statistical operations) o Ex. An inch is an inch, the same amount of difference Interval is the most common type of measurement (typically best) Lacks a meaningful zero point, zero doesn’t mean the complete absence of a property, its just next to one and above minus one We are able to change the location of zero on the scale With an interval level scale, we cannot talk about ratios of raw scale values Can talk about ratios of difference between values Ratio scales are the highest level or measurement Has a meaningful zero point: the complete ant total absence of a property Ratio has everything above AND a zero point You can talk about ratios of actual scale values We can do everything and talk about ratios of actual scale values (if we had them) Rescaling multiplicative ones of the form y=bx, which transforms the S only. These simply stretch or compress the size of the units without changing the location of the 0 point -Exceptions to Stevens’ scales Ordinal categories with meaningful zero but lacking equality of units (e.g. in job analysis, 0=not performed, 1=sometimes, 2=frequently, 3=all the time) o Ordinal scale with a meaningful zero but not scale “Absolute” scale of frequencies/counts (e.g. person has heard voices on 5 occasions) in which no transformations are legitimate -Scaling methods Subject centered: goal = locate people on the scale of measurement (dimension) Stimulus centered: goal = locate stimuli on a scale -Scale metrics are arbitrary Numbers that we assign to people do not in themselves convey truly useful information regarding the amount of the characteristic Ratio scales come close, by virtue of knowing what a zero means (absence) Categorical/nominal scales may, if the names convey sufficient information (e.g. religion, psychiatric diagnosis) -How to infer meaning Even ratio scales are limited once we move to nonzero ratings, as the scale (i.e. relative size of units) can be freely adjusted without changing the basic measurement properties Have to use methods outside the scale to attach meaning to scale score values Normative methods are the most common solution to this problem We need to infer meaning from our scales (Normative scale) how your score stands in a reference group of scores. -Reliability (book version) Reliability refers to the consistency of measurement Types of reliability o Interrater reliability: degree of agreement between two independent raters o Test-retest reliability: degree to consistency over time o Internal consistency reliability: degree to which the items of a measure are in agreement Reliability and validity are two things we would like to see our measurement instruments possess -Sources of test variance Desirable: i.e. due to “true” construct score Undesirable o Enduring characteristics of examinee o Temporary characteristics of examinee (sick) o Systematic characteristics of administration (people administer differently) o Systematic characteristics of test itself (bias) o Chance (guessing) Conceptually: The concept or reliability is the degree to which a test is measuring the underlying trait that we are trying to measure and the relative degree to which it is free of measurement errors -Classical test theory (CTT) has been popular for about the last century Developed around the turn of the century, largely unchanged since Extremely simple, offers a very straightforward definition or reliability Observed score = Xit (for person i at time t) True score = Ti (for person i) Error of measurement = eit (for person i at time t) Model: Xit = Ti + eit Error is basically random, and it is just as likely to help you as it is to hurt you -Reliability (conceptually) Definition: o Degree of freedom from measurement error o Percentage of total variance that is “true” bar. Thus, rxx = s^2T/S^2x = S^2T/S^2T+S^2e With z-scores o When Se^2 = 0, rxx = 1 o When Se^2 = 1, rxx = 0 -Ways to estimate error variance In theory, rxx = correlation between parallel tests: o True part stays constant o Error part changes randomly o Therefore, only true part is available to correlate with other variables, or self o Relative amount of error present attenuates the r between parallel forms Name coefficient of equivalence Reliabilities only range from 0 to 1 Attenuation = to make smaller Parallel tests aren’t really real and we cant actually get them -Test-retest approach Use test as its own parallel form (give test twice) Coefficient of stability: how stable is the test over time? Argument: why could be more parallel than itself? Inter-test interval (ITI): amount of time that elapses between repeated administrations of the test o The longer time elapses the less likely tests are to correlate with themselves o True score is not changed, only the error of measurement, but people change over time (so not really true) o If there is a short interval then you could remember the test and make sure you get the questions right the second time you take it making it inaccurate o One of the better ways to estimate reliability if we are sure the true score is not changed -Test-Splitting methods Coefficient of internal consistency: how homogenous are these subtests? Split-half method: o Break test into 2 parts, correlate them o Odd-even split, first half-second half Advantages: o Don’t need to separate parallel form o Don’t need to repeat administration How similar the tests items are to each other when given at the same time o Split the test in half and compare the two half’s of the test o Drawbacks: if the half’s aren’t even, underestimates the reliability. How homogeneous/similar are the two half’s of the test. The tests is half the length o Longer tests are always more reliable than short tests -Spearman-Brown Prophecy Estimate rxx – for a shorter, or longer, test o May want to see how many more items would be needed to improve rxx to a given standard o See how much rxx would decrease if you shortened the test o rSB = krxx/1+[(k-1)rxx] -Alternatives to split-half Kuder-Richardson formula 20 (KR-20) o Tries to compensate for chance in selecting splits for split- half o Is mean of all possible split-half rxx values Coefficient alpha o Is effectively based on the average of items on the test, give you a measure of the internal consistency of the test -Estimating SEM (Standard error of measurement) Se = SEM = Sx square root of (1-rxx) For z scores, even easier (Sx is equal to 1) Use to set CIs (confidence intervals) around observed score o Are x% sure that true sore lies in the interval o Plus/minus 1 SEM = 68% o Plus/minus 2 SEM – 95% How much error is in the score SEM and reliability are inversely correlated Large values of reliability are desirable, small values of SEM are desirable -Problem of difference scores D = X-Y Reliability of D is a concern: is a function of o rxx (higher is good) o ryy (higher is good) o rxy (higher is BAD) – we don’t want the variables to have a high reliability Bad part: typically with find strong r between X and Y -Validity A scale is valid if it measures what it is supposed to measure Validity also refers to how well a scale predicts other variables (e.g. an IQ test is likely to be reasonably valid predictor of grades in school o When used this way, the scale is call the predictor measure and the measure predicted is called the criterion
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'