1Class 7Conceptualization and Measurement 2Steps of Drawing Causal Graphs1. Identify all variables– Determine the independent var. and the dependent variable.– Is there an intervening variable (mechanism)?– Is there a confounding variable (common cause)?2. Use arrows to show presumed causal order.3. Determine whether the causal relations are positive or negative.Indep. Var.Dep. Var.Intervening Var.Confounding Var.Dichotomous Variables• Dichotomous variables have only two categories. A special case of categorical variables.• Examples: sex (female, male); attitude (agree, disagree); race (white, non-white).• It is appropriate to calculate the mean and variance of dichotomous variables. E.g., Code the two categories as 1 and 0, and we can find out proportions from the mean values of dichotomous variables.24 Million Adolescents Smoke: Or Do They?1. How is “current smoker” defined? How is “heavy smoker” defined?2. Why is the operationalization of “current smoker” not suitable for measuring prevalence of adolescent smoking?4 Million Adolescents Smoke: Or Do They?66.022.534.130-3447.417.136.11837.311.330.31735.19.827.91618.44.323.41515.32.415.7147.90.911.4130.00.06.612% current smokers who smoke daily% dailysmoker% current smokerAgeAdolescents4 Million Adolescents Smoke: Or Do They?• What does this figure say about the relationship between the quantity and the frequency of smoking?• What do you expect the figure to look like for adult current smokers?3Quality of Measurement• We want our measurements to be both precise and accurate.– Precision: the level of details– Accuracy: true to the actual value• Corresponding technical terms– Reliability –Validity Reliability and Validity• Reliability means that a measurement yields consistent scores when the phenomenon is not changing.• Validity means that a measure measures what we think it measures.biasTypes of Validity• Face Validity: The measure makes sense “on its face.”• Content Validity: The measure covers the full range of the concept’s meaning.4Types of Validity• Criterion Validity: The measure matches well with a more direct measure or an existing validated measure.• Construct Validity: The measure is related to a variety of other measures as predicted by theory.Checking for Reliability• Test-retest method – take the same measurement more than once.• Inter-item reliability – expect high association among a set of questions measuring the same concept.• Alternate-forms reliability and split-half method – give slightly different questions to randomly divided sample.• Inter-observer reliability – have two coders working independently.Test-Retest Method• Test-retest method – take the same measurement more than once. • In a panel design, you measure the same quantity twice. Let us call the first measurement y1, the second y2. • Higher correlation between y1and y2indicates higher reliability.• This assumes that the true quantity does not change over time.5Use a Set of Questions• Use multiple items (set of questions) to measure the same concept. (inter-item reliability)• Example of measuring prejudice against blacks.• How much do you agree or disagree with the following statements (1 strongly agree --- 4 neutral --- 7 strongly disagree):1. Black people tend to be lazy.2. Black people tend to be rich.3. Black people tend to be intelligent.4. Black people tend to be violent.5. Black people tend to prefer to live off welfare.6. Black people tend to be patriotic.• Higher correlations among the multiple items indicate higher reliability.Use Alternate Forms• Use alternate forms of the same questions, and check if responses are similar.– Change the order of the questions.– Reverse the order of the response choices.– Slightly change the wording of the questions.• Split-halves method: randomly split survey sample into halves and administer alternate forms of the questionnaire. • If the alternate forms yield similar results, the measurements are reliable.Check Inter-Observer Reliability• Assign two research workers to code the data independently.• Compare their codes.• Higher correlation between the two set of codes indicates high reliability.6Some Measurement Issues • Sensitive Questions• Social desirability• Interviewer effect• Reporting error• Coding errorConsequences of Measurement Errors – Univariate Statistics• Bias = Expectation of measured value – true value• Measurement errors directly affect univariate statistics. – If the measurement of X is unreliable, then the variance of X is larger than it should be.– If X is measured with a bias, then the mean of X is not the true mean.Consequences of Measurement Errors – Bivariate Relationships• When a variable is poorly measured, we may still detect bivariate relationships. • However, the magnitudes of bivariate relationships are usually not correctly estimated.• Example: Gender and church attendance. Suppose that both men and women over-report church attendance by 30%, what happens to estimate of gender difference in
View Full Document