2 Research Objective Show direct cause & effect Study relationships among variables for existing groupsExplain outcomes after the fact
3 Type of Design True Experiment Quasi-Experiment Cross-Sectional LongitudinalExplanatory Case StudyExploratory Case Study
4 Descriptive Cross-Sectionals Many cross-sectionals are purely descriptive of the characteristics of a population, often of a single population with no comparison groupDon’t use one of these in this class; it is not acceptableI will not address these designs further in this class; deVaus’s comments are sufficient
5 Explanatory Cross-Sectionals Explanatory cross-sectionals identify the relationships among several (sometimes many) variables within one or more populationsMost use two or more comparison groups, although some use test how well two or more theories explain the relationships within a single groupThere are generally numerous predictor variables based on the constructs in one or more theoriesThere may also be more than one outcome variable
6 Distinctive FeaturesSelect a representative sample from a population or populationsGroups or populations are defined based on the independent or predictor variables, not groups we create through setting up an intervention or treatment and comparison or control groupWe then measure the independent or predictor and outcome or dependent variablesAssess the strength of the relationships among variables through qualitative or quantitative analysis
7 Did you select one or more characteristics to define the population(s) of interest? YesNoNot an explanatory cross-sectionalDid you select a (preferably statistically) representativesample of each population of interest based on the characteristics used to define the population(s)?YesNoNot an explanatory cross-sectionalDid you collect information each sample based on the predictor and outcome variables of interest?YesNoNot an explanatory cross-sectionalExplanatory Cross-sectional Design
8 Simple Cross-Sectional with Two or More Comparison Populations 2 or more existing comparison populationsA single measurement at one point in timeIn the diagram P=population, RS=representative sample, M=measurement of variablesDefine P1 RS from P1 M1Define P2 RS from P2 M1Define P3 RS from P3 M1
9 Example of Simple Design Research question: How does time between marriage and birth of first child affect marital satisfaction of women?
10 Population & Comparison Groups Required Characteristics: Married women (not divorced or separated; women who have been married only once; and women who have been married less than 10 yearsThese are screening criteria and you use them to try to control other factors that could affect marital satisfaction that are not of interest (previous marriages)Comparison populations/groups: P1 consists of women who were married less than 2 years when the first child was born; P2 is women who were married 2 or more years when the first child was born
11 HypothesesPrimary hypothesis: Marital satisfaction will be lower for women married less than two years when the first child was born (one-tailed).Other theory-based hypotheses: There would probably be several of these, all based on constructs in the theory of this formThere will be a positive correlation between …….Non-theory-based hypotheses: These deal with various characteristics that could affect the relationship between time to birth of first child and marital satisfaction, such as size of the woman’s family, educational level, profession, financial resources
12 Sample & MeasureSelect a statistically representative sample of each population (or, given all the caveats about that, as close as you can get)Measure for all independent or predictor and dependent or outcome variables (theory-based and other)
13 Analysis All analyses can be either statistical or qualitative Test the primary hypothesis first. If there is NO difference between groups for the outcome, you can treat the population as a single population and single sample. However, I recommend continuing testing as two or more populations.Then test the other hypotheses.Most researchers want to create some sort of statistical or qualitative model of the relationships for the comparison groups.
14 Cross-Sectional with Repeated Measures in Time Most cross-sectional designs have no time component, but a time component can be added with repeated measuresIn cross-sectionals, sampling occurs before each measurement – unlike longitudinal designs the participants at each measurement are different people, a different sample, drawn from the population(s)Define P1 RS1 from P1 M RS2 from P1 M2Define P2 RS1 from P2 M RS2 from P2 M2Define P3 RS1 from P3 M RS2 from P3 M2
15 Multiple Cohort Designs Multiple cohorts sampled over time allow you to determine how the relationships between variables changes over time, the assumption being that such changes are due to broad societal processesThe cohorts are simply different age groups in many casesE.g., Financial security over time for people born in , 70-79, etc. In these cases, the cohorts are essentially the theoretical populations. In this example, we might assume that the financial turbulence since 2000 has affected these cohorts differently.P1 (60-69) RS from P1 M1P2 (70-79) RS from P2 M1P3 (80-89) RS from P3 M1P4 (90-99) RS from P4 M1
16 Multiple Measures with Cohorts In other cases, researchers combine multiple cohorts with multiple measures over time. In our example this would reveal how financial security changes over time for people born in the different decades. We might hypothesize that financial security will decline for all populations over time and that the decline will be greater for P1 than P2, P2 than P3, etc.Again, remember that we draw a new sample for each measurement.P1 RS1 from P1 M1 RS2 from P1 M2P2 RS1 from P2 M1 RS2 from P2 M2P3 RS1 from P3 M1 RS2 from P3 M2P4 RS1 from P4 M1 RS2 from P4 M2
17 Cross-Sectional Before & After Some Event These are almost always retrospective designs and are subject to all the limitations of retrospective designs, which include things like people’s ability to remember past feelings and actions, events coloring or changing how we perceived things prior to the event.However, sometimes we get an opportunity that we just cannot ignore and we use these.More commonly, people try to treat these as some form of “quasi-experiment.” They are NOT a quasi-experiment. The “event” was not a planned or intentional intervention. NO intentional poke = NOT an experiment.
18 ExampleAn example might be why people did (or did not) vote for a woman for president (Hilary Clinton). The two populations would be those that voted for her (in the primary) and those who did not.We could use several of the gender theories for this study and I won’t lay all that out here.Assume we did measurement 1 right after the election. We could add a second measurement now, which would improve our study because we could see if the reasons given for voting for her or not remain stable (remember Gersten from last module; his ideas apply here, too).P1 (yes) RS from P1 M1 RS2 from P M2P2 (no) RS from P2 M1 RS2 from P M2
19 SamplingBoth statistical and theoretical generalization depends almost entirely on adequacy of the sampleUsually (not always) need a statistically representative sample with regard to the characteristics of the population that we think may affect the variables (independent and dependent) under study or – given the difficulty in getting a true statistically valid sample, something close to it (review Bernard and Wretman readings from module 6)
20 Describing the SampleDescribe the sample fully in the results section of your study – the reader must have good information to decide if your sample is adequate or not.DeVaus calls this “descriptive analysis,” but it’s not really analysis at all, just description. I have no idea why he uses this term. I have never heard it before.
21 What to IncludeThe norm is to include demographic data about the sample to show whether it is or is not broadly representative of the population(s) of interest with regard to these characteristics.But focus on the characteristics of the sample that may affect the outcomes of the study – that’s what really matters in terms of the adequacy of the sample.Review Gersten in particular (module 11). All of his comments about procedures you can use to establish whether two comparison groups are equivalent in important ways can be used in cross-sectionals. For example, you could use multiple forms of data collection similar to multiple pre-tests.
22 Take Home PointsCross-sectionals are not a substitute for experiments or any other design.They fail to meet all the requirements required to demonstrate direct cause and effect (see Shadish).If well designed, they may provide tentative indications of causality.They allow the researcher to explore relationships among several theoretically related variables, which can add significantly to explanatory power.
23 Cross-sectionals depend on the use of comparison groups to indicate causality. All one group designs are basically descriptive in nature because there is no way to know whether absence of the condition of interest produces the same outcome as presence of the condition without a comparison group.They depend heavily on a representative, independent sample of each comparison population, preferably a statistically representative sample in many cases.However, they are over-utilized in much social research, probably because they are relatively quick and inexpensive – especially the single population descriptive study.