lecture 8 * diagnostics*

look at the values of each of the components

order them form highest to lowest

if a > c > b > d is the order = a good test

1. define test ( purpose)
2. choose scaling method : quantify responses to test
3. itam construction and item analysisi
4. revision
5. cross validation

only with clear , precise purpose and goal

measuremnt goal determines scale type

what you don't want to asses

target popluation

theoretical justicifaction

operationalization

added value

collection of items to which the response is scored and combined into a scale score

1. nominal
2. ordinal
3. interval
4. ratio

categorize ( dog/cat ) but not both mutually exclusive

natural ranking low to high

scale points equidistant

0 point meaningfull thus distance to 0 also menaingfull

parametric : we can make calulations

1. ration
2. interval

non parametric :

1. ordinal
2. nominal

→ we want all of our measurments to be in the same direction

1. stimulus centered
2. subject centered
3. criterion centered

investigate opinion regarding a topic

investigate individual differences

to make adequate predictions

1. expert ranking → criterion centered
2. thurstobe scale → stimulus centered
3. absolute scale → subject centered
4. likert scale / rational scale → subject centered
5. guttman scale → stimulus or subject centered
6. empirical scale → criterion centered

1. expert ranking
2. empirical scale

1. thurstobe scale
2. gutteman scale

1. linkered scale
2. absolute scale

**Glasgow coma scale ( ordinal scale )**

minor 13/ 15

moderate 9 /12

severe less than 8

**Thurstone scale (1929 )**

used for mapping opinions or attitudes

1. develop items in line with construct
2. experts arrange items in 11 categories
3. how much item represents that attitude
4. determine M and SD per item
5. select items : preferably with low SD
6. you have negative and positive statements : which ones do you agree with ?
7. test score of items which person agrees

cathegorize items based on absolute deviation from one reference

1. items and asses multiple age groups
2. items recieve an ' item : difficulty score ' compared to reference ( passed by 50% of the reference group )
3. individual score = z score relative to reference group

tells us something about how above or below the individual is in comparison

score of the individual itself

3 to 7 scales

equidistant scale : same difference between options

extent to which respondent agrees with statemnt

Linker scale

based on science

sensible scoring items

somewhere between ordinal and interval

need to recode

sum of all items

1. create items and asses a large group
2. remove items that show low / negative correlation with total score

Guttman scale (1947)

assumes unidimensionality : respondants agrees to all statements that are less extreme, disagrees with statements that are more extreme

1. items ordered by experts in degree to which they represent the construct
2. ithems that do not meet unideminsionality are left out

Beck's depression inventory

i don't feel particularly guilty

i feel guilty a good part of the time

i feel guilty most of the time

i feel guilty all of the time

not based on theory or experts , just empirical findings

people with schitzofrenia vs people without schitzofrenia

dichotomous items (yes/no)

score = number of items with "yes"

an opinion or attitude

differences betweeen people

an accurate prediction

the optimalization of item quality ( modify , delete, add items )

applicable to all tests

item characteristic curve

1. difficulty
2. discriminatory power
3. reliability
4. validity

p1

0 and 1

a really difficult item

0.3 and 0.7

describes relationship between value of the charcteristic and likleyhood of a correct answer

IQ and the likelyhood to respond correctly

50 % PEOPLE DID ITEM 2 CORRECT

0% OF PEOPLE DID ITEM 3 CORRECT

chace succes level ( 4 questions 25% )

pi = 1+g / 2

pi : item difficulty index

g = chace success level ( when guessing )

median between chace success level and 100% correct

when we have a questionaire with 2 alternative ( chance succes .5)

pi = .75 ( halfway 0.5 and 10)

when we have 4 alternatives ( chance succes 0.25)

pi= .63

if the item difficulty is the highest for one correct answer : best to keep

if the item has the same chance of being picked as the correct answer : take it out , something is not clear

if the item ahs a way too high chance of being picked : way to easy

if anther item is mor elikely to be chosen that the correct one. : take it out ( too hard )

if one of the inccorect items has too chace ( is basically like having a two choice questionnaire not 3)

di

item discrimination index , item characteristic curve

di tells us how effcicntly does an item discriminatioe between person who obtained high and low scores on the entire test

di

does not matter if you gace IQ of 120 or 50 , you have the same chance of responding correctly

di = discriminatory power item

Uc = percentage in upper range who answers corrcetly

Lc =percentage in lower range who anser corrctly

correct answers more frequently chosen by upper range

negative dicriminaroy item : correct answers more frequently chosen by lower range

0 item does not disciminate

poitive and the highest possible

between .3 and .6 is a good item dicriminatory

the upper and lower IQ rage in terms of z score

hiest di = steepest

lowesr di = least steep

item reliability index

riT = orrelation between item and resto of the test

SDi = standard deviation of the item

higher is better = more varince

higher is better



ric = correlation between item and item score