710 likes | 963 Views
Data in the Classroom CSU Fresno November 1, 2010. Presenters. John Korey Cal Poly Pomona, Political Science jlkorey@csupomona.edu Ed Nelson CSU Fresno, Sociology ednelson@csufresno.edu. Workshop Agenda. Introductions (Ed Nelson) SSRIC (Ed) Data for this workshop (John Korey)
E N D
Presenters • John Korey • Cal Poly Pomona, Political Science • jlkorey@csupomona.edu • Ed Nelson • CSU Fresno, Sociology • ednelson@csufresno.edu
Workshop Agenda • Introductions (Ed Nelson) • SSRIC (Ed) • Data for this workshop (John Korey) • Issues and examples • Experimental design (John) • Sampling and Statistical Inference (Ed) • Causality and contingency tables (Ed and John) • Fun with graphics (John) • Change over time (John) • Where can we get the data? (John) • What are we doing this year at Fresno State? (Ed) • Evaluations
SSRIC Social Science Research & Instructional Councilhttp://www.ssric.org
The Council • Oldest CSU affinity group -- founded in 1972 • Each campus has a representative • Works to provide access to data • Promotes use of data analysis in research and teaching
The Council • Annual student research conference on April 29 at San Jose State University • Sponsors attendance at the ICPSR summer workshops in Ann Arbor, Michigan • http://www.ssric.org/participate/icpsr_summer • Works with the Field Institute -- selects faculty fellow (12 questions) – proposal due April 15
Datasets for This Workshop • Based on SPSS for Windows 16.0: A Basic Tutorial (http://www.ssric.org/trd/spss16) • General Social Survey (GSS) 2006 Subset • Based on Introduction to Research Methods (http://www.csupomona.edu/~jlkorey/POWERMUTT/index.html) • American National Election Study (ANES) 2004 Subset • GSS Cumulative File Subset • ANES 2000-2002-2004 Panel Study Subset • U.S. Senate
Issues and Examples • Experimental design • Sampling and statistical inference • Causality and contingency tables • Fun with graphics • Change over time
Design Requirements • Experiments • Random assignment to groups • Manipulation by experimenter of independent (predictor) variable • Quasi-experiments
Types of Experiments • Laboratory • Field
Laboratory Experiment: Prisoner’s Dilemma HOMICIDE DIVISION INTERROGATION ROOM A HOMICIDE DIVISION INTERROGATION ROOM B
Laboratory Experiment: Prisoner’s Dilemma INTERROGATION IN PROGRESS DO NOT ENTER
Laboratory Experiment: Prisoner’s Dilemma JACK’S BAIL BONDS “I’ll get you out if it takes 20 years.” 909/869-461924/7
Field Experiments Gosnell (1927) Gerber and Green (2000)
Resources • The Center for Experimental Social Science
Experimental Design in Survey Research • Telephone vs. face to face (2000 ANES) • Question wording: • Do you favor or oppose doing away with the DEATH tax? • Do you favor or oppose doing away with the ESTATE tax?
Results(2002 ANES) • Favor abolishing “death tax”: 74.3% • Favor abolishing “estate tax”: 71.5% p = n.s.
What do we want to make sure our students understand? • Populations and samples • Parameters and statistics • Sampling variability • Margin of error • Confidence intervals and confidence levels
Basic principle • Samples vary • What factors influence sampling variability? • Size of sample • Population variability • How sample was selected
Using Simulations to Teach Statistical Inference • Draw repeated random samples • Compute sample statistic • Construct chart showing the distribution of these sample statistics • Demonstration – see http://constats.atech.tufts.edu
Estimators and Estimates • An estimator is the method and an estimate is the numerical result • Demonstration – see http://inspire.stat.ucla.edu/unit_09/teaching_tips.php
Resources -- Exercises • Rolling dice and flipping coins – see http://www.causeweb.org/repository/StarLibrary/activities/andrews_2003/ • M&M’s – see http://www.ropercenter.uconn.edu/education/assignments/polling_basics.pdf • Drawing cards (Aces to Kings) – Xuanning Fu (CSU Fresno)
Resources – Web Sites • Roper Center -- Fundamentals of polling: http://www.ropercenter.uconn.edu/education/polling_fundamentals.html • American Association for Public Opinion Research – more on polling -- http://www.aapor.org/Poll_andamp_Survey_FAQs.htm • Sample size calculator -- http://www.surveysystem.com/sscalc.htm
What do we need to do to establish cause and effect? • Statistical relationship • Causal ordering • Eliminate alternative explanations
Example • Religiosity and how to regulate the distribution of pornography – data set – gss06_subset_for_classes_modified2.sav • RELITEN – how religious the respondent is • PORNLAW – how the respondent feels about regulating the distribution of pornography
Spuriousness • Are there any alternative explanations (other than the causal one) for the relationship? • Can we think of any alternative explanations for RELITEN and PORNLAW? • Gender might account for this relationship. Women are more religious than men and also more likely to want to restrict the distribution of pornography • In other words, the relationship between X and Y might be spurious. So what we need to do is to test for spuriousness
Testing for Spuriousness • Independent variable (X) is RELITEN • Dependent variable (Y) is PORNLAW • Control variable (C) is SEX
Conclusions • We found out that the relationship of RELITEN and PORNLAW was not spurious when we controlled for SEX • But does that mean that we can conclude that the relationship is never spurious? • What does this say about proving causality?
Applying this to the Classroom • Start with examples that make sense to students • Move to examples with real data that students can run • Generalize to issues of testing causality • Can show that a relationship is not causal (i.e., it’s spurious) • Can never prove that a relationship is causal.
Example: Specification Open General Social Survey Subset Does level of education influence the relationship between political views and party identification?
Specification (continued) From Menu bar, go to:Analyze Descriptive Statistics Crosstabs Dependent variable (first box): partyid Independent variable (second box): polviews Control variable: (third box): degree Statistics: Kendall’s taub Cells: Column percentages
Specification (continued) Look at pattern of Kendall’s taub statistics
Example: Reactivity We know that the race of the interviewer in face-to-face interviews affects what people tell us about race We know that the perceived race of the interviewer in telephone interviews also influences what people tell us What about the gender of the interviewer in face-to-face interviews?
ANES Example Open anes04s We’ll going to use three variables GENDER – gender of respondent INTGENPO – gender of interviewer WORKMOM – do you agree or disagree [that a] working mother can establish just as warm and secure a relationship with her children as a mother who does not work? Let’s start by using the gender of the interviewer (INTGENPO) as our independent variable and WORKMOM as our dependent variable
ANES Example Continued What did we discover? Respondents interviewed by women are more likely to agree that working mothers can have a warm relationship with their children Now let’s see if this is true for both male and female respondents. Let’s control for GENDER – gender of the respondent We discover that it is true for both men and women. It appears that the gender of the interviewer does influence what people tell us about working mothers and their children
ANES Example Implications Since about 75% of the interviewers in this survey were women, this has some serious implications. This suggests that we will overestimate the percent of people that feel that working mothers can have a warm relationship with their children
Box and Whiskers Plots • Open senate file (senate_mod.sav) • Compare acu and dwnom scores • Graphs Legacy Dialogs Boxplots Clustered Summarize by Separate Variables Define • 1st box: acu, dwnom; 2nd box: party; 3rd box: name; OK
Box and Whiskers Plots (continued) • Convert acu and dwnom to Z scores • Analyze Descriptive Statistics Descriptives • Move acu and dwnom to right window • Check Save standardized values as variables
Box and Whiskers Plots (continued) • Compare Zacu and Zdwnom scores • Graphs Legacy Dialogs Boxplots Clustered Summarize by Separate Variables Define • 1st box: Zacu, Zdwnom; 2nd and 3rd boxes remain the same; OK
Sample Size and the “Margin of (Sampling) Error” http://www.surveysystem.com/sscalc.htm
Just the Facts http://pollingreport.com/guns.htm
Poll Aggregators http://www.pollster.com/polls/
Do It Yourself Prognostication http://uselectionatlas.org/PRED/