Psychology 340 Exam 1

Inference

A test is a measurement device or technique used to quantify behavior or aid in the understanding or prediction of behavior.

Is a specific stimulus to which a person responds overtly

Scales relate raw scores on test items to some theoretical or empirical distribution.

• Measure current behavior
• Predict future behavior
• Infer hidden behavior

• States-temporal; change
• Traits-fixed

• The individual
• A group

• Ability test-measure previous learning
• Aptitude test-measure potential for acquiring a particular skill
• Intelligence test-measures intelligence

• A person's general ability to...
• solve problems
• adapt to changing circumstances
• think abstractly
• profit from experience

Chinese civil service test

The American government established the American civil service commission

Developed mental test

Developed to educate and evaluate the mentally handicapped.

A group given the test under standard conditions.

A sample that represents the group we are using comparisons.

How you perfomed compared to others who took the test.

• Stimulus
• Response
• Projective
• Ambiguous
• Face-Valid

• Inkblot
• Could not find meaningful standardized responses
• Still widely used
• Full of Shit...lol

uses empirical methods (factor analysis) to determine of a test response

• Concise descriptions of lots of quantitative information.
• demographic breakdown
• prevalance of behavior

• Deduce things you cant observe directly.
• Measure something in a small group of ppl.- the sample
• Infer the results apply to a larger group of ppl. - the population

using rules to assign numbers to objects

Magnitude-moreness

Equal intervals-difference between any 2 points on the scale is the same

Absolute zero- it is possible to have none of the quality measured

Nominal-not really scales, assigning random designator numbers to ppl or things

Ordinal-have moreness but not equal intervals or absolute zero. E.g. putting ppl in order shortest to tallest

Interval- have magniude and equal intervals but not necessarily absolute zero. E.g. temperature

Ratio- have magnitude, equal intervals and absolute zero

Displays how many times (frequency) each score (distribution) was obtained on a scale.

• x-axis=each score, from lowest to highest
• y-axis= # of times each score was obtained

Increased sample size; bell-shaped curve

• pos.=tails to the right
• neg.=tails to the left

IQ scores have a pos. skew

The distance between two consecutive measurements in your distribution.

Indicate the particular score below which a defined percentage of scores falls.

Divide the percentage scale into four groups

Divide the scale into tenths

Divide the scores into standard 9ths

this creates 3 groups= the lower, middle and high group

The middle 50% of scores (25%-75% percentiles)

The average score in a distribution

mean= (sum of all scores) / (# of scores)

The point at which 1/2 the scores are above and 1/2 are below

The "bump"

most common score

The variation of the scores around the mean

The average deviation around the mean

calculation: SD=Square root of the variance

Indicate how far from the mean a score is

universal expression of SD

• scores above the mean are + z-scores
• scores below the mean are - z-scores

between -3.0 and +3.0

The mean did not equal 0, as it is with z-scores it equaled 50

standard deviation did not equal 1, as it is with z-scores it equaled 10

Mathematical translation; transformations standardize the distribution.

• Average performance by standardization sample
• The standardization sample sets the norms

• Compare individuals to a normative group
• E.g. class exam scores

Compare an individual to a criterion on a specific skill, tasks, or knowledge

• A measure that can have multiple values
• E.g. weight, test score, etc.

As on changes, another changes

• Correlation
• Regression
• Multiple regression

A picture of the relationship between 2 variables

One variable increases on the horizontal, or x-axis

One variable increases on the vertical, or y-axis

Help us make predictions between scores of two variables

Like a slope line on a graph

• steep= x may predict y
• flat= y may predict x

a= y-bx

The difference between the actual score and the predicted score

the best fitting line minimizes the residual

correlation is like regression, except scores from each variable are standardized

Describes the magnitude and direction of relationship between 2 variables

• Number...
• -1.0 - +1.0

• Pos.- variables 1 and 2 go up or doen together
• Neg. - variables 1 and 2 move in opposite directions
• No correlation

The statistics used to measure correlation

Assume that there is no relationship between 2 variables

2 countinuous variables

yes/no; true/false; pass/fail; etc...

All dichotomous variables are categorical

No Pearson's r for dichotomous variables

Between one continuous and one dichotomous variable

• The standard deviation of the residuals
• Predicts fit of the regression line
• Smaller is better

• The correlation coefficient squared (r²)
• Indicates how much of the variation in Y is due to X

Using the regression equation from one group of subject to predict performance in a different group of subjects

The amount of decrease in predictability from cross-validating

If the variability of a variable is extremely restricted, significant correlations may be difficult to find even if they are there

Creates factors: groups of related variables

Is all about error.

• Difference in true ability and measurement of ability.
• The inevitable inaccuracy of our measurements.
• How much our test do not reflect reality.

• find the magnitude of error for each test
• try to minimize error for all tests

reliable

measure tangible things

• measure intangible things
• one that may over or under estimate the measurement

• Influences how we think about and calculate reliability
• assumes each person has a true score

• We do not report the true score
• we report the observed score

Bell-shaped

Is accurate

Is inaccurate

• Time-sampling
• test-retest method
• Item-sampling
• parallel forms method
• split-half method

• Administer the test at 2 different times
• compare each test-takers scores

• Caveats:
• stable traits
• carry-over effects

when the variance is equal on different parts of the test

Dichotomous test & Likert scales

α alpha

σT²

σx²

• Uses a sample of items, not the entire domain of items
• pick several items from your original test
• give that small test
• record observed scores
• estimate true scores fro these observed scores
• More items = more reliability (generally)

Test that adapts to the test taker

• Subtract one score from another
• same test given at 2 points in time
• 2 sub-scores from one test

Make the comparison in Z-units

• The error of a difference score is inflated
• It absorbs error from both scores in the difference score.

Calculate RDS if you know... α of each test and r between these test

Even tests with high α and r may have low RDS

The similarity of rater 1 and rater 2's measurements

• K = Pr (a)-Pr (e)
• 1-Pr (e)

• -1.0 - 4.0