Class 16Class OutlineContent AnalysisSteps of Content AnalysisExample: Who Wrote the 15th Book of Oz? Baum or Thompson? – Jose BinongoSlide 6Principle Component Analysis (PCA): A Statistical Method for Dimensional ReductionThe Attribution of the Royal Book of OzStrengths of Content AnalysisWeaknesses of Content AnalysisAnalyzing Existing StatisticsSecondary Data SourcesHistorical and Comparative ResearchSlide 14Macro-Causal AnalysisMethod of AgreementExample: Causes of Revolution in Russia, China, and FranceMethod of DifferenceExample: Causes of Revolution in France and GermanyWeaknesses of Macro-Causal AnalysisClass 16Unobtrusive ResearchClass Outline•Two General Types of Research Methods–Reactive/obtrusive•Surveys•Experiments•Field research–Unobtrusive•Content Analysis•Analyzing Existing Statistics•Historical and Comparative AnalysisContent Analysis•A research method for systematically analyzing and making inferences from text.–Texts are broadly defined as books, chapters, essays, interviews, discussions, newspaper headlines, articles, historical documents, speeches, conversations, advertising, etc.•Allows for both qualitative and quantitative operations.–Qualitative – involves coding and categorizing text and identifying relationships among constructs identified in the text.–Quantitative – statistically analyzes coded text.Steps of Content Analysis1. Identify a population of documents or other textual sources for study.2. Determine the units of analysis.3. Select a sample of units from the population.4. Design coding procedures for the variables to be measured.5. Test and refine the coding procedures.6. Base statistical analyses on counting occurrences of particular words, themes, or phrases, and test relations between different variables.Example:Who Wrote the 15th Book of Oz?Baum or Thompson? – Jose BinongoPopulation = 29 books of OZ: 14 by Baum, 14 by Thompson, and The Royal Book of Oz (the 15th book)Variables are frequencies of function words.50 variables are used, corresponding to the 50 most frequently used function words. Units of analysis are text blocks of 5000 words.N = 223.Analysis dataset is a table of 233 observations by 50 variables.Units of analysisPrinciple Component Analysis (PCA): A Statistical Method for Dimensional Reduction1st PC2nd PCInterpretation: “which,” and “that” are on the opposite of “up,” “on,” “down,” and “over” on the 1st PC. This means that if a text block uses a lot of “which” or “that,” it tends to use fewer position words. In this application, the 1st PC coincides with authorship.position wordsThe Attribution of the Royal Book of Oz•Each data point is a text block of 5000 words, either by Baum or by Thompson.•PCA is a multivariate statistical technique for dimensional reduction. In this application, it preserves the variation of the 50 variables by using only 2 dimensions.•The first dimension distinguishes authorship.•Baum’s texts are clustered on the right side of the 1st Pc, while Thompson’s are clustered on the left side of the 1st PC.•The Royal Book of Oz is clearly written by Thompson.ThompsonBaumStrengths of Content Analysis•Looks directly at communications via texts or transcripts. Richness of data.•Economy of time and money.•It's simple to repeat (a portion of) the study if necessary.•Permits you to study processes occurring over a long time.•Unobtrusive - Researcher seldom has any effect on the subject being studied.•Possible to achieve high reliability.Weaknesses of Content Analysis•Limited to the examination of recorded communications.•Is often devoid of theoretical base. Tends to simply consists of word counts.Analyzing Existing Statistics•Existing statistics can be the main source of data or a supplemental source of data.•Problems with validity - Often existing data don't cover the exact question.•Reliability is dependent on the quality of the statistics.Secondary Data Sources•U.S. Bureau of Census (Data retrieval system American FactFinder) http://www.census.gov •Bureau of Labor Statistics http://stats.bls.gov •Data Archive – ICPSR http://www.icpsr.umich.eduHistorical and Comparative Research•Seeks to discover patterns in the histories of different cultures by comparing two or more cases across space and/or time. •Main resources for observation and analysis are historical records.•Where do history and sociology intersect and differ? – Nomothetic -- prevailing mode in sociology.– Idiographic -- prevailing mode in history.•Examples:–Durkheim’s study of suicide–Weber’s study of capitalismHistorical and Comparative ResearchCaution:•Documents and other evidence may have been lost of damaged.•Be wary of bias in data sources.–Available evidence may represent more newsworthy figures.–Written records will be biased toward those who were more prone to writing.Macro-Causal Analysis•Purpose # 1 - developing new explanations.•Purpose # 2 - “setting scope conditions” for theory.•Method of Agreement•Method of Difference•CommentsMethod of Agreement1. Select cases with the same outcome (dependent variable).2. Compare possible causal factors across cases.3. Try to isolate one or a few features that are the SAME across cases.4. Conclude that this is the causal factor producing the outcome.•Example from Skocpol’s States and Social Revolutions.Example: Causes of Revolution in Russia, China, and FrancePossible CausesRussia China FranceDifferences: type of government, religion, level of industrialization, etc.State breakdownYes Yes YesPeasant revoltYes Yes YesRevolution Yes Yes YesMethod of Difference1. Select cases with different outcomes (dependent variable).2. Compare possible causal factors across cases.3. Try to isolate one or a few features that are DIFFERENT across cases.4. Conclude that this is the decisive difference (i.e. the causal factor) producing different outcomes.•Another Example from Skocpol’s States and Social Revolutions.Example: Causes of Revolution in France and GermanyPossible Causes France Germany in 1848Fragmented ElitesYes YesMilitary Threat Yes YesVillage autonomy high, so peasants could revoltYes NoSuccessful revolutionYes NoWeaknesses of Macro-Causal Analysis•Researcher must assume deterministic causality.•There might be multiple causes or interaction effects.•Unlikely that you could measure all causal factors.•Selection on the dependent
View Full Document
Unlocking...