Conceptualization
[Link] [Link] [Link] [Link],Dimension,and Indicator [Link]
Definition
Conceptualizationistheprocessofdevelopment andclarificationofconcepts. Inotherwords,clarifyingone'sconceptswith wordsandexamplesandarrivingatprecise verbaldefinitions. e.g.,whatismeantbyeducation? Amountofknowledgeandtrainingacquiredin school."
AnotherExample
Whatdowemeanby"socialstatus?
Wealth(millionaire) Prestige(Harvardprofessor) Power(militarygeneral)
Thesearecalleddimensions ofsocialstatus.
Dimensions
Weclassifydifferentmeaningsintodifferent [Link]"dimensions." Aconceptmayhavemorethanone dimension(e.g.,asincaseofsocialstatus). Atapracticallevel,weareusuallymore interestedindimensionsthaninconcepts (whicharemoreabstract,vague).
Indicators
Whenadimensionisnotdirectlyobservable,we useindicators. Forexample,tomeasurepower,wemayuse
(1)numberofpeopleunderyoursupervision (2)extentofyoursupervision(workrelatedonly, orsleepandfood?) (3)yourannualbudget (4)amountofequipmentunderyourcontrol
ComparisonofConcept,Dimension, andIndicator
Inpractice,thethreetermsareoften interchangeable(e.g.,gender,race). Onedifferenceisthelevelofabstraction:
ConceptDimensionIndicator HighlyabstractAbstract Concrete
Oneconceptmayhavemultipledimensions; andonedimensionmayhavemultiple indicators
ARelatedterm:Variables
Avariableisastatisticalterm,meaninga quantitythatcantakeondifferentpossible values. Bothdimensionandindicatorcanbe variables. Whenaconcepthasonlyonedimensionwith oneindicator,aconceptispractically equivalenttoavariable.
Summary
Concept
D2
D1
D3
...
I12
I11
I13
...
Example:SES
SES
Wealth
Prestige
Power
I31
I32
I33
...
Measurement
Whenitcomestomeasurement,weare talkingaboutvariablesandindicators. Definitionofmeasurement:"theassignment ofnumbersorlabelstounitsofanalysisto representvariablecategories." Numbersmeandifferentthingsunder differentcircumstances.
TypesofMeasurement
Nominal measurement Ordinal measurement Interval measurement Ratio measurement
Nominal Measurement
Nominalmeasurementisasysteminwhich casesareclassifiedintotwoormore categories onsomevariable. Arbitrarynumericalassignments.e.g.,
Race=1forwhite,2forblack,and3forAsian.
Twocriteriaforclassifications:
Exhaustiveness Mutualexclusiveness
Ordinal Measurement
Inordinalmeasurement,numbersindicatethe rankingorderonadimension. e.g.,foratypicalattitudequestiononsurveys,
Strongly Agree 5 Somewhat Agree 4 Neutral 3 Somewhat Disagree 2 Strongly Disagree 1
There is no intrinsic scale you only know relative rankings. Should not do arithmetic (such as averaging).
Interval Measurement
Intervalmeasurementassumesequal distancesorintervalsbetween"numbers." Numbersrepresentnotonlyrankingsbutalso values. e.g.,(70,90,80)is thesameas(80,80,80) for thetest component ofyourSoc.357finalgrade.
Ratio Measurement
Ifanintervalvariablehasanabsolutezero,it becomesaratiovariable. E.g.,weight,numberofsiblings,birthrate, etc. Comparethreetemperatures:C,F,andK.
AComparisonoftheFour
Information Provided Classification Rank order Equal intervals Relative value Nominal X Ordinal X X Interval X X X Ratio X X X X
More general
More specific
Qualityofmeasurement
[Link] Precision istheextenttowhichnumerically detailedmeasurementremainsmeaningful (e.g.,measuringannualincome). [Link] Accuracy istheextenttowhichthe measuringinstrumentmeasureswhatitis intendedtomeasure.(Commonlygaugedby reliability andvalidity).
Consequenceofmeasurementerror: descriptiveversusexplanatorystudies
Inexplanatorystudies,weareinterestedin relationshipsbetweentwoconcepts,ortwo [Link],the relationshipisstillthere. e.g.,genderandreligiosity. Saytheresearchinterestliesinthedifferential [Link] theabsolutelevelby30%forboth,westillhave anaccuratemeasureoftherelativevalue.
GenderandReligiosityExample
Gender Men Women True 50 100 Measured 35 70
i.e.,explanatoryanalysisisrobust tomarginal changesofmeasurements. Reason:overestimationorunderestimationforall unitsofanalysis. Bewareofdifferential under(over)estimation.
MethodsforAssessingReliability
Testretestmethod Asinpaneldesign,youmeasurethesame [Link] measurementy1,thesecondy2. Reliabilitymeasure:correlationbetweeny1 andy2. Lowcorrelationmeanslargerandomnoise.
MethodsforAssessingReliability (Continued)
Splithalfmethod Youmaycalculatecorrelationsamong differentitemsonthesamesurvey instrument,assumingequivalenceofdifferent items. Highcorrelation:highreliability
Bias(LackofValidity)
Inregressioncontext, yi = + xi +i i [Link],it couldbethat i =4+i ,wherei iswellbehaved(i.e.,hasa meanofzero).Thus,4isthebias.
MethodsforAssessingValidity
[Link] Seewhetherameasurementmakesanysense toyouandtoothers(alsocomparewith similarindependentmeasures GSS) [Link] Doesthismeasurepredictothermeasures thatcanbemeasuredmoreobjectively. Usecorrelationor regressiontechniques.
Example:PredictabilityofSATscoreof CollegePerformance
IfSATisagoodmeasureofacademicability,why doesntitpredictcollegeperformancewell? Problems: (1)selectionbias; (2)mediatingfactorsbetweenfirstmeasurement andcriterion(e.g.,instruction); (3)measurementproblemsforthecriterion(e.g., gradeinflation,differentialgrading,etc.).
MethodsforAssessingValidity (Continued)
[Link] Theextenttowhichanempiricalmeasurement reflectsaspecificdomainofcontent. [Link] Constructvalidityisconcernedwiththeextent towhichaparticularmeasurerelatestoother measuresinwaysconsistentwiththeoretically derivedhypotheses.
AGraphicModelforConstructValidity
Ix1
Ix2
Iy1
Iy2
