Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
74 Cards in this Set
- Front
- Back
Increasing internal validity is best achieved by:
random selection matching random assignment blocking |
Random Assignment
|
|
Internal Validity
|
a study is said to have
INTERNAL VALIDITY when it permits the conclusion that there is a causal relationship between the IV and DVs. in other words, when there aren’t extraneous variables that might explain the observed scores |
|
8 threats to internal validity
|
1) history
2) maturation (any internal change in Ss during experiment) 3) testing 4) instrumentation 5) statistical regression 6) selection 7) differential mortality 8) experimenter bias |
|
1) Most powerful method of increasing internal validity
2) Other methods of increasing internal validity |
1) Random Assignment
2) Matching (grouping by status on extraneous variable then randomly assigning) Blocking (studying effects of extraneous Ss characteristic (eg IQ) by making it another IV Holding extraneous variable constant ANCOVA (mathematical adjustment to data so that Ss are equalized in terms of status on extraneous variable. |
|
External Validity
|
a study has external validity to the extent that results can be generalized to other settings, times, or people.
|
|
Threats to external validity
|
1) interactions b/w selection & tx
2) interactions b/w testing & tx 3) interactions b/w hx and tx 4) demand characteristics 5) Hawthorne Effect 6) order effects (in a repeated measures study) |
|
Strategies to increase external validity
|
1) randomly selecting Ss from population
2) conducting naturalistic or field research 3) using single- or double-blind designs 4) counterbalancing (if order effects are an issue) |
|
Quasi-experimental design
|
a manipulable variable is studied but Ss are NOT randomly assigned (often b/c Ss are in pre-existing, intact groups like hospital wards, classrooms)
|
|
Correlational Design
|
variables aren’t manipulated and no causal rx is assumed, e.g., assessing rx b/w gender and IQ score
|
|
What are the input and output variables called in correlational studies?
|
Input = PREDICTOR variable
Output = CRITERION variable |
|
3 types of developmental research designs
|
1) longitudinal
2) cross-sectional 3) cross-sequential (Ss of different age grps are studied over a short period of time |
|
Time-Series Design
|
the DV is measured several times at regular intervals both b/f and after a tx. Helps control threats to internal validity, but can’t control for history effects.
|
|
Factorial design
|
a study with more than one IV
e.g., IV#1) 3 different levels of a tx IV#2) 2 levels of symptom severity |
|
Stratified random sampling
|
taking a random sample from each of several subgroups of the total target population to ensure proportionate representation of the defined population subgroups.
|
|
Cluster sampling
|
in this sampling strategy, the unit of sampling is a naturally occurring group of individuals, rather than the individual. Eg., if target pop is city residents, cluster sampling might involve breaking city into 30 square blocks, randomly selecting blocks, then interviewing all individuals in those sampled blocks.
|
|
Latin Square Design
|
a counterbalancing design in which the ordering of txs across different conditions allows administration in every possible order.
|
|
What is the strength of internal validity in correlational research?
|
Correlational research doesn’t have any internal validity, because it cannot infer causation.
|
|
Time Series Design
|
give multiple pre-test measurements over time, compare to the multiple post-test measures given over time. Because you’re giving a series of tests over time pre- and post-, if you observe a decrease in post-test scores you can rule out practice effects or maturation because you would have observed the same declines in the pre-test series.
|
|
What is ANCOVA and what is it for?
|
ANCOVA
a stats method for increasing internal validity by adjusting DV scores so that Ss are equalized in terms of their status on 1 or more extraneous variables. |
|
Single-subject designs don’t fare well when…
|
there’s lots of variability in the target variable at baseline.
|
|
Protocol Analysis (in Qualitative research)
|
analyzing verbatim reports from subjects, typically when they think aloud as they are performing a task.
|
|
Nominal
|
unordered categories
(e.g., gender) |
|
Ordinal
|
ordered like ranks, but not equidistant
(e.g., likert scale ranks) |
|
Interval
|
successive points are equidistant, but no true zero.
You may add and subtract, but not multiply or divide (e.g., IQ) |
|
Ratio
|
successive points are equidistant WITH true zero.
All math functions (+ - x / ) are ok (e.g., height) |
|
Positively skewed test
|
positive tail (e.g., a hard test with few high scores, mostly low scores)
Positive Skew = Positive TAIL |
|
Negatively skewed test
|
negative tail (e.g., an easy test, with few low scores)
Negative Skew = Negative TAIL |
|
By convention, small greek letters are used to represent ________, while roman letters are used to represent ________
|
GREEK = POPULATION VALUES
ROMAN = SAMPLE |
|
How do you calculate variance?
|
Variance = SD²
the SQUARE of the standard deviation |
|
How do you calculate SD?
|
it’s the
SQUARE ROOT of the VARIANCE |
|
What’s a linear transformation?
|
LINEAR TRANSFORMATION
a transformation of scores that does not distort the distribution’s shape e.g., conversion from raw → z-scores doesn’t change the shape of the distribution |
|
T-scores
|
T-Scores
50=mean every 10 points = 1 SD therefore, a score of 60 is one SD above the mean (and would equate to a z-score of +1.0) |
|
A _________ is referenced to items on the test, whereas a ________ is referenced to other scores in the distribution
a) percentage score b) percentile rank |
A (a) PERCENTAGE SCORE is referenced to items on the test,
whereas a (b) PERCENTILE RANK is referenced to other scores in the distribution |
|
Is transforming raw scores to percentile ranks a …
a) linear transformation b) non-linear transformation and WHY? |
b) non-linear b/c the distribution of percentile ranks is by definition flat, and unless your raw score distribution is flat, you’ve changed the shape of the distribution, making this a NON-LINEAR transformation.
|
|
What’s the difference between a statistic and a parameter?
|
statistics are about sample values
parameters are about population values |
|
Standard Error of the Mean
|
an index of the expected inaccuracy of a sample mean.
i.e., the expected deviation between an estimated population mean and the true population mean. SD/ √(N) |
|
If the standard error of the mean is 2, what does that actually mean?
|
a standard error of the mean = 2 means that the sample mean can be expected to deviate from the actual population mean by 2 points either way.
|
|
Type I error
|
rejecting a true null hypothesis
saying there IS a rx when there really ISN’T TYPE I = ALPHA (usually at .01 or .05) |
|
Type II error
|
retaining a false null hypothesis
saying there ISN’T a rx when there really IS. TYPE II = BETA |
|
Power
|
the probability that a statistical test is able to detect a true effect of an independent variable
|
|
Parametric test
vs. non-parametric test |
PARAMETRIC
designed for interval or ratio data NON-PARAMETRIC designed for nominal or ordinal data |
|
As sample size INCREASES, standard error of the mean __________
a) increases b) decreases c) doesn’t change in a predictable way |
b) decreases
|
|
One-tailed vs two-tailed hypotheses
|
ONE-TAILED: we expect the experimental mean will be different than the control mean in a particular direction
(e.g., a reading intervention will IMPROVE reading scores) TWO-TAILED: we expect the experimental mean will be different than the control mean, but we don’t know in which direction. |
|
What are the pre-requisites, or assumptions, when using parametric tests?
|
Parametric Tests
1) normal distribution of DVs 2) homogeneity of variance (variance of DV of all groups in the study is equal) 3) Independence of observations* (each score is independent from other scores in the group) – the most important assumption of parametric tests |
|
What’s a ONE SAMPLE t-test used for?
|
comparing a sample mean to a known population mean
|
|
What a t-test for INDEPENDENT (e.g., randomly assigned, uncorrelated) samples?
|
to compare the means obtained from 2 independent samples
|
|
What’s the t-test for CORRELATED SAMPLES (e.g., matched samples, pre-test/post-test samples)?
|
to compare the means of two correlated samples (such as a before and after)
|
|
What test is used to determine difference between means when a study has one IV, one DV, and 3 or more groups?
|
a ONE-WAY ANOVA
which yields a F score Mnemonic: always think of 1-ANOVA-3: One-way = one IV. 3 = at least 3 groups. |
|
What does an ANOVA tell you, and what doesn’t it tell you?
|
ANOVA (one way)
is used to determine difference between means when a study has one IV, one DV, and MORE than 2 groups tells you if there is a significant difference among the samples’ means DOESN’T tell you precisely which means are significant different. usually post hoc tests are used to pinpoint the exact patterns of the means |
|
What test is used when a study involves two or more IVs and one DV?
|
a FACTORIAL (AKA eg “2-way”, “3-way”) ANOVA
allows for assessment of both main effects AND the interaction effects of the two IVs together on the DV. |
|
What test is used to analyze data with multiple DVs and AT LEAST one IV.
|
MANOVA
|
|
What is the advantage of using a MANOVA rather than multiple separate ANOVAs?
|
reduces the probability of a Type 1 error.
|
|
What are the 4 non-parametric tests?
|
1) chi-square
2) Mann-Whitney U 3) Wilcoxon Matched-Pairs 4)Kruskal-Wallis Test |
|
What non-parametric test is used to compare observed frequencies of observations in nominal categories vs. frequencies expected under the null hypothesis?
|
Chi-square
|
|
What non-parametric test is used to compare 2 independent groups on a DV measured with rank-ordered data?
|
Mann-Whitney
|
|
What is the non-parametric alternative to the t-test for independent samples?
|
Mann-Whitney
|
|
Of the post-hoc tests you might use after, say, an ANOVA found significant difference b/w your group means, which one is most conservative – that is, protects against Type 1 errors?
|
The Scheffe
|
|
Which post-hoc test would you use after an ANOVA if you only want to do pairwise comparisons?
|
Tukey
|
|
What test is used to adjust DV scores to control for the effects of an extraneous variable?
|
ANCOVA
|
|
What is a regression for?
|
REGRESSION = PREDICTION
when 2 vars are correlated, it estimates the value of a “criterion” (outcome or “predictee”) var on the basis of scores on a “predictor” (input) variable. e.g., given a student’s GRE scores, what will their GPA be? |
|
What is the “multiple” in multiple regression referring to?
|
“multiple” refers to multiple predictor (input) variables used to predict scores on one criterion (outcome) variable.
|
|
Discriminant Function Analysis is used for?
|
DISCRIMINANT FUNCTION ANALYSIS
a corelational technique used when the goal is to classify individuals into groups based on their scores on MULTIPLE predictors |
|
Partial Correlation is used for?
|
it’s a correlational technique used to assess the rx b/w TWO variables with the effects of another variable “partialled out” (statistically removed)
|
|
What is point biserial correlation used for?
|
correlations between
one continuous variable (i.e., intervally or ratio scaled) and one DICHOTOMOUS variable (e.g., gender) e.g., what’s the correlation between income and gender? |
|
What’s a biserial coefficient used for?
|
it’s a correlation test for two continuous variables in which one is made artificially dichotomous
e.g., what’s the correlation between exam scores and income level when exam scores are split into “high” and “low”? |
|
What is Phi correlation used for?
|
a correlation statistic used when both variables are dichotomous (naturally)
|
|
What is a tetrachoric coefficient used for?
|
a correlation test used when both variables are artificially dichotomized.
|
|
Spearman’s Rho is used for?
|
Spearman’s Rho (aka Rho)
used to correlate 2 variables that have been ordinally ranked e.g., 2 judges rank the same set of observations and their agreement would be determined by a Rho. |
|
Eta is used for?
|
Eta
Correlation for relationships that are NON-LINEAR! e.g., correlation between anxiety and performance which is U shaped |
|
What is multiple correlation used for?
|
MULTIPLE CORRELATION
predicts/estimates scores on a criterion using the scores on MORE THAN 1 predictor variable higher “multiple Rs” (multiple regression value) mean a stronger rx b/w predictor and criterion e.g. predicting college GPA using the predictor vars of HS GPA, SATs, and IQ. |
|
What are
forward and backward STEPWISE multiple regressions? |
FORWARD stepwise:
starts with 1 predictor and adds predictors 1 at a time. With each addition, you calculate how much predictive power is gained. The predictor w/ largest correlation w criterion is retained and you continue adding until no further increase in predictive power is gained. BACKWARD stepwise: starts w all potential predictors, removing one at a time. When you get to point were removing predictors significantly decreases predictive power, you stop removing and keep remainders for the final equation. |
|
What is CANONICAL correlation?
|
CANONICAL correlation is like a multiple correlation for cases of
multiple predictors and multiple criterion variables. Relates 2 or more predictors to 2 or more criterion vars in 1 statistical analysis |
|
What is DISCRIMINANT FUNCTION ANALYSIS used for?
|
DISCRIMINANT FUNCTION ANALYSIS
combines scores on 2 or more variables to determine if they can be used to predict CRITERION GROUP MEMBERSHIP e.g., using several IQ tests to a group of children to predict who will belong to a high achieving vs low achieving group |
|
Logistic Regression is used when?
|
LOGISTIC REGRESSION
used to make predictions about which criterion group a person belongs to used instead of Discriminant analysis when: a) assumptions aren’t met (not a normal distribution of scores, not homogeneity of variance) b) predictors = are categorical/nominal data. |