WARNING: These are preliminary notes. Final notes are subject to change at any time, so always revisit this webpage just before each class.

TOPIC ONE: INTRODUCTION TO THE COURSE

ACCESSING YOUR DATA FILE (classroom)

0) Remember that to log into the computer, you must type in "mspoll" and "shaffer" at the two places provided. Also, click on Advanced, and in the Context block, type in "Classes.Sociology.Bowen.CAS.MsState", and be very precise about the upper and lower case letters.

1) Access your statistical analysis package first by double clicking on SPSS from the menu

2) When the screen reads "SPSS for Windows". Double click on the option, "MORE FILES".

3) When the screen says, "Look in... SPSS," click on the top right arrow, pull the top side bar down, and double click on "NETWORK NEIGHBORHOOD"

4) You will need to double click on a series of subdirectories. When you double click on the first, the next one will come up:
"BOWEN2"
"ARCHIVE"
"AREAS"
"ARCHIVE"

5) Now press the bottom right arrow, and get to "MISSISSIPPI POLL." Double click it.

6) Now double click on the file name, "Mpoldat".

TOPIC TWO: PHILOSOPHY OF THE SCIENTIFIC METHOD

Classical era- 700 BC to 1850 AD

Philosophical, ought, how should things be, justice, who rules, obligations of citizens and government.

Institutional era- 1850 to 1900 AD

Traditional approach, process, how a bill becomes law, structure, legal approach, unitary state as a rational actor, descriptive, historical.

Traditional approach to analysis combines the classical and institutional eras.

Transitional era- 1900 to 1945

Behavioral era- 1945 to present- characteristics are:

1) Science, theory, predictions, explanation, patterns:
Examples:
a) Theories of presidential voting behavior- voters are influenced by their party identification, issue attitudes, and evaluations of the candidates.
Hypothesis 1: Voters psychologically identifying with the Democratic Party are more likely to vote for a Democratic presidential candidate, compared to voters identifying with the Republican Party.
Hypothesis 2: Voters who are liberal on policy issues are more likely to vote for the Democratic presidential candidate, compared to voters who are conservative on issues.
Hypothesis 3: Voters who prefer the personal qualities of the Democratic presidential candidate are more likely to vote for the Democratic candidate, compared to voters who prefer the qualities of the Republican presidential candidate.

1b) American voting behavior across offices (president, U.S. Senate, U.S. House, governor): vote is influenced by party identification, ideological issues, incumbency, name visibility, presidential coattails.
Hypotheses 1 and 2 are similar to the preceeding hypotheses.
Hypothesis 3 (non-presidential races): Voters living in districts or states with Democratic incumbents are more likely to vote for the Democratic candidate, compared to voters living in districts or states with Republican incumbents.
Hypothesis 4 (non-presidential races): Voters who can recall only the name of the Democratic candidate are more likely to vote for the Democratic candidate, compared to voters who can recall only the name of the Republican candidate.
Hypothesis 5 (non-presidential races): People who vote for the Democratic presidential candidate are more likely to vote for the Democratic congressional/gubernatorial candidate, compared to people who vote for the Republican presidential candidate.

1c) Congressional roll call votes are affected by party of congress member, urbanization of district, blue-collar presence in district, and region. (Over 80% of variance in four scales--economic, welfare, agriculture, civil liberties--is explained).
Hypothesis 1: Democrats are more likely to have liberal roll call voting records on economic and welfare type issues, compared to Republicans.
Hypothesis 2: Congress members from urban districts are more likely to have liberal voting records on economic and welfare issues, compared to Congress members from rural districts.
Hypothesis 3: Congress members from districts with a high percentage of blue collar workers are more likely to have liberal voting records on economic and welfare issues, compared to Congress members from districts with a low percentage of blue collar workers.
Hypothesis 4: Congress members from outside of the South are more likely to have liberal voting records on economic and welfare issues, compared to Congress members from the South.

d) Patricia Jernigan's MSU dissertation examines race-sex composition changes of southern state and local governments since 1975 by job type, as well as salary differences across sex-race groups; then explains why African-Americans are underrepresented or make lower salaries than whites, such as less college education than whites, or few black elected officials pressuring agency heads for diversity.

2) Data gathering and research are theory directed
Examples:
a) For presidential voting behavior, national survey of voters was conducted, asking their party identification, their attitudes on public issues, and their likes and dislikes of the major party candidates.
b) Voting behavior across offices used existing national surveys on party identification, ideological self-identification, voter ability to recall the names of the non-presidential candidates, party of the incumbent, and the reported vote for four different offices.
c) Congressional roll call votes (Clausen study) divided roll calls into four types of issues; it identified party of the congress member, used demographic information about the member's district.
d) Jernigan dissertation has a wealth of information for 15 southern and border states, at 5 time points, for 8 occupation groupings, for four sex-race groupings. Hence, data collection focuses on very specific information, such as median salary for each sex-race groupings for all 600 cases, and numbers of workers for all 600 cases which are converted into percentages of workers in each race-sex grouping. Black officials, education operationalize.

3) Value free
Examples:
a) All three of the above studies seek to predict how people will vote, and explain why they vote the way they do. Whether the researcher is a liberal or conservative, Democrat or Republican, obviously has no impact on their research approach or findings.
b) Specific presidential findings: Reagan won in 1980 because of dissatisfaction with Carter's leadership and the bad economy, not because voters loved Reagan's conservative philosophy; Clinton won in 1992 because of the recession, not because voters preferred more liberal policies such as gays in the military and nationalized health care.
c) Researchers were not biased when they pointed out that the Democrats were the majority party in party identification until 1980, since Democrats nearly always controlled Congress; when the party identification gap between the parties started closing during the Reagan years, our surveys pointed out this Democratic party loss.
d) Liberal professors may personally hope to find continued racial discrimination in southern state governments, but Jernigan data may document an increased African-American presence in state governments over time, plus continued black underrepresentation due largely to educational differences in workforce rather than agency discrimination.

4) Interdisciplinary- sociology, psychology, economics
Examples:
a) The earlier American presidential election studies of the 1940s relied heavily on sociology, proposing that group membership affected the party voted for outside of the South. Rurality, Protestantism, and higher income predicted more Republican votes, while urban residence, Catholicism, and lower income predicted more Democratic votes.
b) My study of Balance Theory drew on psychology. People tend to acquire and retain psychologically consistent beliefs and attitudes. If a person likes a candidate, they tend to believe that the candidate agrees with their own positions on issues regardless of whether the candidate actually does; if a voter dislikes a candidate, they tend to believe that they are in disagreement with the candidate on the issue.
c) Shaffer worked with economics professor (Chressanthis) in studying whether U.S. Senate election margins were accountable to the public, which they were in an indirect sense. Elections were affected by presidential coattails, campaign spending, divisive primaries, and preceding election margin.
d) Jernigan study borrows from: sociology regarding the impact of black elected officials on white agency heads' receptivity towards diversity hiring; economics, regarding how to define labor pool of eligible workers, such as college education, and economic barriers for African-Americans to acquire a college degree.

5) Methodological sophistication-
Examples:
a) Shaffer study of balance theory relied on the 1994-1996 American national panel study to examine cognition change over time.
b) Shaffer and Chressanthis study of Senate accountability used pooled time-series, cross-sectional approach. All even-numbered years from 1976 thru 1986 were included, as were all 33 state contests in each election year. Regression and probit were used.
c) Shaffer voter turnout decline study used regression to ascertain that turnout had declined between 1960 and 1976 because of: growing independence from party identification; decreased political efficacy; decreased newspaper use; growing young and old population.
d) Jernigan study uses SPSS statistical package; objective data from EEOC on diversity and salaries among state workforces; percentages and regression.

6) Individual and group level of analysis
All of the studies cited above used the individual as the unit of analysis--either the individual voter, or the individual congressman.
Some studies examine the group, focusing on such subjects as interest groups, congressional committees, demographic groups, etc.
Jernigan study used group of unit of analysis--she compared black females and black males to white females and white males in terms of number of state agency workers and median salaries of these four groups.

Criticisms of Behavioralism- are people and events predictable, can we be value free

TOPIC THREE: SCIENCE AND THE HISTORY OF THE DISCIPLINE OF PUBLIC ADMINISTRATION

Read two assigned articles from PAR

Lee article: Political Science, Public Administration, and the Rise of the American Administrative State.
Late 1800s in U.S. exhibited social problems of modern industrial society. Political scientists turned to science, whose truth and objectivity would provide guidance to political action and institutional change which would be above class and group interests. Reforms backed by public administrators: a British-style executive-led system with a weakened Congress resulting in efficiency; civil service reform which provided expertise and neutrality; municipal reform with professional city managers.
From 1900 thru 1930. Scholars distrusted a public which was intolerant of civil liberties exercised by unpopular groups. They sought quantification and the use of statistics, since statistical trends would reveal objective laws and regularities that were independent of human existence and above ideological disputes. "Like natural science, social science as knowledge for prediction and control may be used for technical intervention and engineering of social and political processes." Science rested upon mathematical precision, prediction and control would provide its social value, and knowledge produced would be objective and neutral and value-free. Franklin Roosevelt's New Deal was example of effort for social control and management through the use of scientific knowledge. Focus on strong leadership and centralized political power--a strong state to cope with social problems. FDR's term saw rise of Executive Office of President, and rise of presidential power.
PA Discipline today: PA scholars must have a strong research methods background in order to conduct high-quality research and publish in top journals such as PAR. Scientific method permits an accurate and objective description of the real world. Discipline is committed to such politically liberal values as diversity of workforce, which includes African-Americans, women (family leave policies necessary), the disabled (ADA act), and much research focuses on these subjects.

The Troublesome Cleft: Public Administration and Political Science.
Some public administration (PA) scholars argue that PA is a separate discipline, distinct from political science (PS). PA has its own national association, for instance, and at some universities it has a separate department. Click here to view ASPA's website. Doctoral students should become members of ASPA and receive PAR journal; student membership is only $35. Authors argue that PA can learn much from political science discipline.
1. Political science offers more rigorous scientific training than PA. While PA often relies on case studies, PS benefitted from behavioral era of 1950s National Election Studies research. PA dissertations have been criticized for being weak quantitatively and too focused on case studies; need for more scientific approach to research. MSU response is to require additional methods courses, and have political scientists teach them.
2. Political science concepts such as power, justice, conflict, and policy are relevant to PA. PA workers need to recognize power relationships among their elected superiors. Iron Triangle is a relevant concept. Justice is illustrated by social equity of public policy. Taxation equity. Casino dissertation done by Rivenbark finds that lower SES have more access to gambling when they live in a county where it is legalized; problem of lottery and casino gambling being a regressive tax. Projects distributed across city wards of different racial composition. Conflict resolution: Fordice versus Black Caucus. Policy: how political and economic factors affect it; is state policy funding affected more by wealth of state or by amount of partisan competition.
3. PS and PA both focus heavily on individual level of analysis. Many PA articles focus on diversity in workplace, efficiency of individual workers.
4. PS has a more theoretical approach, while PA focuses more on practical applications. Yet scientific study at individual level of workers' motivations, values, and attitudes that influence behaviors is important to PA practitioners. Doctoral education is different from practical application education of MPPA program; doctoral education is theory and research oriented.
5. It is very important for PA practitioners to understand their political environment, and how it affects their behavior and operations. A mass public shift toward more conservative attitudes may shape welfare policy, for instance. A new GOP-controlled Congress in 1995 will challenge social welfare agencies' budgets. PA dissertations increasingly include political variables, not merely administrative ones. Jernigan study includes presence of black elected officials as an inducement for state agency heads to diversify their workforce.

TOPIC FOUR: THEORY BUILDING

THEORY BUILDING

Four characteristics of a good theory:

1) Explanation- why does something happen
Examples:
a) Presidential voting model. People vote Democratic because they psychologically identify with the Democratic party, because they are liberal, and because they prefer the Democratic presidential candidate's characteristics.
b) Voting for different elective offices. In the 1994 Senate election between Republican Senator Trent Lott and Democratic challenger Ken Harper, people may have voted for Lott because they knew more about him than they did about the opponent (they could recall his name), because he was the incumbent (and had seniority and experience); also, maybe they were Republican identifiers (GOP has grown in the state), maybe they were conservative themselves, and maybe they rated President Clinton's job performance poor (no presidential coattails in midterm year, so presidential job performance rating used as a new variable).

2) Prediction- if we know people's positions on the independent variables, can we predict their positions on the dependent variables
Examples:
a) In presidential vote model, if a voter is a Democrat, a liberal, and prefers the Democratic candidate's attributes, we predict that they would vote for the Democratic presidential candidate. If a voter is a Republican, a conservative, and prefers the Republican candidate's attributes, we predict they would vote for the Republican presidential candidate.
b) In the congressional vote model, the party of the congress member is a very important predictor. Hence, Democratic congress members might vote in a liberal direction on roll calls about 90% of the time, while Republican congress members might vote in a conservative direction on roll calls about 90% of the time. Examples of Mississippi Republicans.
c) Clinton impeachment vote was very partisan in committee. In House Judiciary Committee, conservative white male Republicans opposed demographically diverse liberal Democrats. Click here for info about the Judiciary Committee members.

3) Generalizability- does theory apply to different situations and circumstances and time and geographic areas
Examples:
a) Presidential vote model. Applies to any time span; 19th century would have different parties (Whigs and Democrats, Federalists and Democratic-Republicans). Can apply to different geographic areas, such as other nations (Ohio State professor Bradley Richardson used party identification model in Japan, Netherlands, Germany, France, Britain, Italy).
b) Cross-office model and congressional roll call model both apply to any time span. Cross-office model can extend to other offices besides governor and congress, such as statewide elective offices and state legislative races.

4) Parsimony- simple with few independent variables, simplest theory is best if everything else is equal
Examples:
a) Presidential Vote model. Is parsimonious, as has only three predictors--party identification, issues, candidates.
b) Cross-office model and congressional roll call model also have limited number of predictors, and are relatively parsimonious.

Independent variable is the predictor; it comes first temporally and causally, it causes the dependent variable.

Dependent variable is the effect, it is being caused by the independent variable.

Ideology -------------> Presidential Vote

(Independent (Dependent Variable)

Variable)

Hypothesis is a statement of a relationship between concepts.

Example: self-identified conservatives are more likely to vote Republican, compared to self-identified liberals.

Hypothesis test- example with crosstabulations, put independent variable at top, dependent variable at the side. Calculate column percents.

VOTE FOR:	LIBERAL	MODERATE	CONSER- VATIVE
CLINTON	78%	58%	28%
DOLE	22%	42%	72%
	100%	100%	100%

Theory

|

|

-Hypothesis-

- -

Concept <------------------------> Concept

(Relation)

This is the theoretical level- general, abstract

Indicator <------------------------> Indicator

(Hypothesis Testing)

Operationalizing your concept is to select specific indicators of your abstract concepts. Hypothesis testing occurs at the indicator level, and it measures the relationship between the indicators.

If hypothesis is rejected, maybe the indicator is not valid.

Religiosity example of a theory.
At the theoretical level, the two principal concepts are Social Deprivation and Religiosity. The principal hypothesis at the theoretical level is that people who are socially deprived are more likely to be intensely religious than are people who are not socially deprived.
Operationalizing the concepts is to choose valid, specific indicators of those concepts. One indicator of religiosity might be frequency of church attendance. An indicator of social deprivation might be annual family income before taxes. The major problem with operationalizing one's concepts is whether the indicators are valid measures of those theoretical concepts. Is a person who attends church twice a week necessarily more religious than someone who never attends church, but who reads the Bible and prays daily? Is a person with a large family income, but who also has a large family size, necessarily well-off financially? Can you think of more valid indicators of these concepts of social deprivation and religiosity?
Hypothesis Testing measures the relationship between the indicators. Are people with low family incomes more likely to attend church weekly, compared to people with high family incomes? Are people with lower net financial worths more likely to pray daily, compared to people with high net financial worths? If your hypothesis is rejected, there may be two reasons. Perhaps your theory is rejected, or perhaps your indicators are not valid measures of your concepts.

RELATIONSHIP BETWEEN FAMILY INCOME AND CHURCH ATTENDANCE (GSS)

RELATIONSHIP BETWEEN FAMILY INCOME AND PRAYER FREQUENCY (GSS)

Groupthink article discussion, is it a good theory?
It explains why disastrous foreign policy decisions occur.
Can it be generalized to other nations and to business and university decisionmaking settings?
Does Groupthink have predictive ability? Can it predict disastrous foreign policy decisions ahead of time, so that we could prevent such disasters?
Is Groupthink a parsimonious theory with few predictors?

Another theory is provided in Miller, Kerr, and Reid's PAR article on Gender-Based Occupational Segregation in Municipal Bureaucracies. Its theory is derived from Theodore Lowi's conceptual framework of three types of policies--distributive, regulatory, and redistributive. This article is an excellent example of testing a prominent theory in our discipline, and was published in the lead journal of the field of public administration.

TOPIC FIVE: INTRODUCTION TO RESEARCH PAPER AND LIBRARY WORK

YOUR RESEARCH PAPER

1) Introduction- discuss the importance of your subject. Discuss your initial expectations. Example of gender gap in presidential voting--why are women voting slightly more Democratic than are men? Why is this subject important? Why do you think this female Democratic bias is occurring?

2) Your model and hypotheses. List all five of your hypotheses, and draw your model.

Example of a model and its hypotheses:
Assume that sex is the earliest, independent variable; presidential vote is the latest, dependent variable; ideology and income are the two intervening variables located between sex and vote.

GENDER.....(H1).......> Ideology .....(H2).....> PRESIDENTIAL
Male or...................(H3)..............................> VOTE
Female.....(H4)........> Income ......(H5)........> (D or R)

The hypotheses are:
H1: Women are more likely to be liberal, compared to men.
H2: Liberals are more likely to vote Democratic for President, compared to conservatives.
H3: Women are more likely to vote Democratic for President, compared to men.
H4: Women are more likely to have lower incomes, compared to men.
H5: Lower income people are more likely to vote Democratic for president, compared to higher income people.

3) Literature review. Need at least 10 academic sources. The articles should be grouped by hypothesis, even if you must discuss the same article more than once. For my on-line bibliography of articles since 1975 in four political science journals, click here.

4) Methods section. Provide information for each of the years of the Mississippi Poll that you are using. For information about the polls, click here. Information on specific years is near the end of that webpage. Discuss each of your indicators, how they are worded and coded, and how valid they are. Discuss the statistical techniques you will use to test your hypotheses, such as crosstabs, regression, etc.

5) Findings-- bivariate. Test each of your 5 hypotheses using crosstabs. Compare percentages using complete sentences, which test your hypotheses. Mention the direction of the relationship, the magnitude of the relation using gamma or average percentage difference, and statistical significance level using chi-squared. Also, draw all tables and provide variable and value labels, and column percents and totals.

TABLE 3

GENDER DIFFERENCES IN PRESIDENTIAL VOTE

Chi-squared < .05
Note: Percentages total 100% down each column.
Source: 1999 Mississippi Poll, conducted by Mississippi State University.

Example of text paragraph:

Hypothesis 3 of my model states that women will be more likely to vote Democratic for president, compared to men. In the 1999 Mississippi Poll, 50% of women indicated that they intended to vote for Democrat Al Gore, compared to only 39% of men who indicated an intended Democratic vote. Hence, the hypothesis is upheld, as women are more likely than men to express an intention to vote Democratic. The magnitude of the relationship is 11%, which is the percentage difference between men and women in their likelihood of voting Democratic. This relationship between gender and the vote is statistically significant at the .05 level, as the Chi-squared statistic is significant at the .04 level.

6) Findings- multivariate tables. At least control for your two intervening variables. Provide information listed in 5. What do these multivariate tables tell you about which of the variables is important in influencing the dependent variable, and about how important each is.

7) Multiple Regression and Recursive Causal Model. Test your model in a recursive causal model format, using at least three separate multiple regression equations. Your path coefficients are Betas. Draw your model with the Betas above each line connecting pairs of variables. Use only Betas that are statistically significant at the .05 level. Discuss whether there are any methodological problems with your regression analyses, such as multicollinearity, regression assumption violations due to dichotomous variables, etc.

8) Conclusions- Redraw your model, discuss your findings and literature, suggestions for future research.

9) References- alphabetize your references by authors' last name. Give full citations for scholarly articles, books, and other citations.

TOPIC SIX: RESEARCH DESIGN

RESEARCH DESIGN

1) Problem Formulation- what are you studying, why is it important. Rivenbark article, casino gambling, importance due to regressivity, hurts poor, addiction.

2) Literature review- thorough. Political science journals are: American Political Science Review, American Journal of Political Science, Journal of Politics, American Politics Quarterly, Public Opinion Quarterly. For a list of on-line political science articles, click here.

Public administration journals: Public Administration Review; check syllabus.

Suggests hypotheses.

3) Identify Unit of Analysis- what are you collecting data on, getting information about what units.

The four units of analysis are: Individual, county, state, nation
a) Individual level examples are public opinion polls.
b) County level example is a public policy study examining spending in each of Mississippi's 82 counties.
c) State level example is a public policy study examining spending in each of the nation's 50 states.
d) Nation unit of analysis example may be relating each of the world's nation's suicide rate to its absence of Catholicism in its population.

Test your ability to identify the unit of analysis of ten different studies by going to the directory for the Masters level methods class, and accessing one of the sample tests for Test 1. Or click here.

4) Design data collection mode- survey, roll call, aggregate (unit analysis above individual), content analysis
a) Survey is a public opinion survey. It can be of the mass population, or of a more specialized group, such as government workers.
b) Roll call mode deals with congressional or state legislative votes on public issues, and often includes demographic characteristics of their district's constituents.
c) Aggregate mode deals with a level of analysis higher than the individual. It deals with cases that combine numbers of individuals, such as counties, states, etc. The data are often secondary data analysis, collected by government agencies.
d) Content analysis is a study of the characteristics of messages, such as how ideologically biased is the mass media, and how many liberal or conservative themes are voiced by a President or governor.

5) Pre-test survey anticipates validity problems with indicators, and suggest variables you left out. For a statewide public opinion poll of 600 Mississippians who are asked 100 questions, you might ask a random sample of 25 Starkville residents the 100 questions, and then ask the interviewers whether the respondents had difficulty answering any of the questions, and if so why.

6) Data collection, surveys use CATI system, or secondary data analysis (use existing dataset).
CATI stands for Computer-Assisted Telephone Interviewing system, and is used for the researcher to collect her own data on an original study. Secondary data analysis relies on existing data sources, such as the University of Michigan National Election Studies conducted every two years, or the MSU Mississippi Poll conducted every two years.

7) Data reduction, usually obsolete with CATI, often needed with in-person and mail surveys; enter data into SPSS program. Classroom demonstration of dataset creation using SPSS.

8) Design statistical analysis technique, do a simple one first such as crosstabs. In Rivenbark dissertation, a simple table showing the percentage of people's incomes devoted to casino gambling, divided by different income levels, was very instructive. He then turned to a complex multiple regression equation explaining amount of income spend on gaming based on demographic characteristics.

9) Perform analysis, get results, show tables and results, discuss results

10) Conclusions- what you found, so what, importance, theory upheld or rejected, future research directions

TOPIC SEVEN: DATA SOURCES

EXAMPLE OF METHODS SECTION IN PAPER

To test the hypotheses in my model, I used the 1998 telephone survey conducted by the Survey Research Unit of the Social Science Research Center at Mississippi State University. A random sampling technique was used to select the households, and a random method was employed to select one individual in each household to interview. Six hundred eight adult Mississippi residents were interviewed from April 14 to April 26, 1998. The response rate was 64%. The sample was adjusted by demographic characteristics (education, sex, race, adults, phone numbers) to ensure that all social groups were adequately represented in the survey. Census data for 1996 were used to obtain population estimates for education, and census data from 1990 were used for race and sex population estimates. With 608 people surveyed, the sample error is plus or minus 4%, which means that if every Mississippi resident had been interviewed, the results could differ from those reported here by as much as 4%.

AGGREGATE DATA (ECOLOGICAL FALLACY)

Ecological fallacy is the incorrect assumption that relationships existing at the aggregate level also exist at the individual level.

Example of religion and presidential vote in the 1940s. Two tables showing individual level relations and aggregate marginal results.

Example from 1990 census- foreign born and college degrees aggregate relationship
STATE.....% FOREIGN BORN.....% COLLEGE DEGREE
Mass................9%......................20%
N.H.................5%......................18%
Vermont.............4%......................19%
N.Y................14%......................18%
N.J................10%......................18%
Alab................1%......................12%
Ark.................1%......................11%
La..................2%......................14%
Miss................1%......................12%
Ga..................2%......................15%
S.C.................2%......................13%

Another 1990 census example- % black and % Republican presidential vote at state level of analysis
STATE.....% BLACK.....% REPUBLICAN PRES. VOTE IN 1988
Alabama.....25%.............59%
Georgia.....27%.............60%
Miss........36%.............60%
Virginia....19%.............60%
Iowa.........2%.............44%
Minn.........2%.............46%
Penn.........9%.............51%
Wash.........3%.............48%
Wisc.........5%.............48%

Example from Joe Parker book on Mississippi electoral patterns

Problems with cross-sectional surveys that gather data at only one time point:
1) Inability to study change
2) Hard to make recursive causal assumptions

Panel design definition: the same people, asked the same questions, at two or more time points. Each time point is called a wave.

Problems with panel designs:
1) Cost
2) Tracking movers
3) Panel mortality, causing biased later samples
4) Measurement conditioning

Examples of panel studies:
1) National election studies panels of 1956-58-60 and of 1972-74-76. The later was able to study the effects of Watergate
2) 1980, 4 wave U.S. national election study. It examined the effects of campaigns on voters.
3) The M. Kent Jennings panel of high school seniors and their parents. Wave 1 was in 1965, wave 2 in 1973, and wave 3 in 1982. Subject was socialization and persistence of attitudes over time.

TOPIC EIGHT: LEVELS OF MEASUREMENT

LEVELS OF MEASUREMENT

NOMINAL- lowest level of measurement, mere classification. No ability to order the categories.
Examples are religion. Use crosstabulations.

ORDINAL - able to order the categories of the variable in terms of a category having more of something than the next category. But can't determine how much more of that quality that the category has compared to the other category.
Example is rating job performance of public officials into excellent, good, fair, or poor categories.

INTERVAL- able to order the categories, and also determine how much of the quality the category has. Usually has numbers that have meaning to denote how much of the quality each category has.
Example is income. Use regression techniques.

Test your ability to classify indicators by nominal, ordinal, and interval levels of measurement by turning to the sample tests, test 1. Click here.

TOPIC NINE: RELIABILITY AND VALIDITY

RELIABILITY

Definition- repeated measurements of a concept (the indicator) should yield similar results.

Tests of reliability:

1) Test-Retest- using the same indicator on the same people at two or more time points. Should have consistent responses at both time points.

TEST-RETEST RELIABILITY TEST OF PARTY IDENTIFICATION

1976 Partisanship

How much stability is there in this table? How many people have given the same response at both time points? Count the number of people in the diagonal. The number remaining stable in attitudes = (9 + 13 + 4 + 5 + 5 + 7 + 6) = 49. The total number of people in the table is 100. Hence, 49% of the sample has remained stable in attitudes. Is 49% high or low reliability? The stable percent must be compared to chance alone. Chance stability is the number of stable cells, divided by the total number of cells in the table. Hence, chance stability is 7 / 49 = 14%. Since 49% is significantly higher than 14%, this indicator is reliable.

2) Alternate Forms (Parallel Forms)- using two or more indicators on the same people at one time point. Should have consistent responses for both indicators.

ALTERNATE FORMS

1999 Party Identification

Consistent responses for both indicators are Democrats who believe that the Democratic party is best for people like themselves, Republicans who believe that the Republican party is best for people like themselves, and Independents who believe that both parties are equally good for people like themselves. The number of consistent responses is (209 + 43 + 128) = 380.

The total number of people is 542. The percentage of people who give consistent responses is:

380 / 542 = 70%. How reliable is the party identification indicator compared to chance alone. Chance is the number of consistent cells divided by the total number of cells: 3 / 9 = 33%. Since 70% is significantly greater than 33%, the party identification indicator is reliable.

3) Split Half- using multiple indicators of a concept on the same people at one time point. Forms two scales with each combining people's responses on half of the indicators. The two scales' scores should be consistent for people.

4) Cronbach's Alpha- used for multi-indicator indexes, calculates how reliable the component indicators are. Ranges from 0 for unreliable to 1 for most reliable.

Reasons for low observed reliability:

1) Non-attitudes, use filter variable.

2) Random measurement error.

3) Actual attitude change, especially with test-retest and great time intervals.

Definition- are we really measuring what we think we are measuring.

Types of validity tests:

1) Face Validity- on its face, it appears to be valid. Simple concepts, such as a ruler. Just use it.

Very well established indicator, don't question it.

2) Construct (Criterion) Validity- relate your questionable indicator to more well established indicators, and see whether it behaves as you expect it to behave.

CONSTRUCT VALIDITY

Questionable Indicator is Party Identification

Note: Cell entries are percentage vote for Republican candidate among each of the seven party identification categories.

Our expectations are that the percentage Republican vote would increase steadily as one moves from the most Democratic party identification category of Strong Democrat to the most Republican party identification category of Strong Republican. Examine the 1988 presidential vote indicator, we see a steady increase in Republican vote as we move from Strong Dem. to Strong Rep. with two exceptions. Only 87% of Weak Republicans voted for Republican Bush, while 94% of Independent Republicans voted for Bush. Those two categories should have reversed percentages, so circle both of those cells, since they involve validity problems with the party identification indicator. Examine the 1996 presidential vote and you find two sets of validity problems among Democrats and Republicans. Circle the four cells having validity problems.

Repeat this validity test for the other five vote indicators, and discuss the validity problems with the party identification indicator that you find.

3) Convergent-Discriminant Validity Test- different measures of the same concept should yield similar results; the same measures of different concepts should yield different results. Examine correlation matrix.

CONVERGENT-DISCRIMINANT VALIDITY TEST

Correlation Matrix of State Spending Preferences (1981-1999)

Note: data are based on the 1981-1999 Mississippi Poll, with some fictitious data included to simplify table interpretation.

Convergent-discriminant validity tests help to determine if your multiple indicators of one concept are actually measuring only one concept, or whether your indicators are measuring more than one concept (a multi-dimensional concept). Generate a correlation matrix as indicated above, and remember that the correlations range from 0 for no relationship to 1 for highest relationship. Then, pick out the highest correlations in order of their size. In the above table, the validity test shows that spending is a multi-dimensional concept involving four separate dimensions (concepts). Those dimensions are: social welfare (poor, day care, health), education (elementary-secondary and college), economic development (industry, tourism), and public order (police, prisons). The environment indicator does relate to any of these four. Hence, any researcher combining all ten spending indicators into one scale that supposedly measures one concept of public support for government programs has validity problems, since there are four dimensions rather than one dimension of state spending.

4) Factor Analysis- can be used as a validity test for testing whether a concept is multi-dimensional.

TOPIC TEN: DIMENSIONAL ANALYSES

Computer work on factor analysis and forming additive scales.

SPSS program (earlier version) of Factor Analysis. Click on Statistics, Data Reduction, and Factor, or their equivalents. Include all variables that you wish analyzed in the variables window. Extraction method is Principal Components, which is the default. Rotation, change to Varimax rotation, so that all dimensions are independent of each other. Examine the Rotated Component Matrix on your computer output--it indicates the correlation between each indicator and the resulting dimensions. In the state spending example, three dimensions (factors) result: welfare, industry-order, and education. You can also save the factor scores as variables, or you can create simple additive scales for each of these three dimensions.

TOPIC ELEVEN: SURVEY RESEARCH--SAMPLING AND SURVEY TYPES

SURVEY RESEARCH

Historic Problems with Polls:
1) Biased Samples- 1936 Literary Digest poll example;
2) Time Bound polls- 1948 Dewey-Truman race, 1980 Reagan landslide;
3) Likely voter problems- Mason-Dixon 1991 Fordice problem.

Sample Error Correlates:
1) Sample Size- larger samples produce smaller error;
2) Homogeneity of Population- united populations produce smaller error;
3) Smaller populations produce less error; for larger areas, one must interview more people;
4) Cluster Samples- produce about 20% higher error.
See textbook, page 170 for a chart on sample size needed to achieve a given level of sample error.

TABLE OF SAMPLE ERROR

HOMOGENEITY OF POPULATION

Note: Cell entries are sample error figures.

Types of Surveys: In-person; Telephone; Mail; Mixed Methods; briefly discuss each.

PROS AND CONS OF SURVEY TYPES

In-person-- pros:
1) Observe and clear up R's confusion
2) Obtain objective information about R's (respondent) lifestyle
3) Visual Aids use
4) Establish rapport? High response rate?

In-person-- cons:
1) Expensive
2) Safety of interviewer
3) Interviewer fraud

Telephone-- pros:
1) Quick
2) Cost effective
3) Centralized interviewing- no fraud
4) Interviewer safety

Telephone-- cons:
1) Excludes those without telephones
2) No visual aids-- voice dependent

Mail-- pros:
1) Cheap
2) Use with specialized population

Mail-- cons:
1) Excludes illiterates
2) Can't control who answers survey
3) Can't control order of questions answered
4) Slow
5) Incomplete forms
6) Low response rate?

Probability Sampling (text, chapter 5). Definition of probability sample: each population unit has some chance of being in the sample, and that chance can be calculated. Types of probability samples:
1) Simple random sampling- each population unit has the same chance of being included in the sample; random number table or generator can be used
2) Systematic sampling- need a list of population units, use a skip interval.
3) Stratified random sampling- ensures the sample adequately represents certain population groups; used if a group is a small proportion of the population, or if researcher wishes to compare groups. Two types of stratified random sampling are:
3a) Proportionate stratified sampling- number of units selected from each stratum is directly proportional to the size of the population in that stratum.
3b) Disproportionate stratified sampling- a larger percentage of units is taken from some strata than from others (usually the smaller strata).
4) Cluster and Multi-Stage sampling- different stages may be PSU (primary sampling unit), city or rural route, 3 city blocks, 5 households, 1 adult in each.

Telephone Sampling Techniques:
1) Telephone directory sampling- problems;
2) Telephone directory sampling, randomize last digit;
3) 2-stage random digit dialing; pure random digit dialing is too inefficient
4) Purchase working telephone numbers from marketing firm

Sampling within the household:
1) Kish method, ask household resident to list first names of all adults, then toss dice to select adult to interview;
2) Carter-Trodahl method: multiple selection tables asking number of adults and number of men in household;
3) Sociological last birthday method; problem that it oversamples women.

Demographic Groups Undersampled in Surveys, especially Telephone Surveys:
1) High School Dropouts
2) Lower income
3) African-Americans, other ethnic minorities
4) Men
5) Young
6) Old

Weighing the Sample:
1) Weight by number of adults in household
2) Weight by inverse of number of different telephone numbers
3) Compare sample and census on demographics
4) Weight by undersampled groups, such as high school dropouts and men; compare sample and census
5) Repeat step 4 until obtain representative weighted sample

TOPIC TWELVE: SURVEY RESEARCH-QUESTIONNAIRE CONSTRUCTION, IMPLEMENTATION

ACTUAL EXAMPLES:

(From Survey Research for Public Administration, by David H. Folz, Sage Publishers)

1) Perceptions of local problems- p. 5, 22, 107
A) No problem, Minor Problem, Major Problem
B) Most serious problem
C) Agree-disagree with problem statements

2) Quality of local services- p. 8
A) Excellent, good, fair, poor

3) Policy preferences- p. 5, 22
A) Single most important change
B) How improve quality of life- not important, somewhat important, very important
C) One policy- oppose or favor, strong or some.

4) Funding priorities- p. 5, 22
A) Single choice, reduce funding first
B) City spending- too little, about right, too much

5) Tax hike backing- p. 20
A) Specific increase for specific policy

6) Citizen usage satisfaction- p. 8
A) Filter question, did they use service?
B) Satisfied or dissatisfied, very or somewhat
C) How often policy met expectations

7) Business usage satisfaction- p. 6
A) Survey gov't workers about complaints heard
B) Survey businesses about specific problems, Overall satisfaction

8) Wording problems- p. 99
A) Loaded or leading
B) Double barreled
C) Too complex, double negative (Miss Poll)
D) Unbalanced alternatives (Blacks treated same as whites or worse)
E) Acquiescence bias (agreement bias)- especially on agree-disagree items
F) Sensitive items- use income categories
G) Social desirability- race items

DISSERTATIONS USING SURVEY RESEARCH
Numerous survey research-based studies have been published in public administration and public policy journals, such as PAR
One MSU dissertation using survey research was Rivenbark's, which studied whether casino gambling in Mississippi constituted a regressive tax, and whether the regressivity was greater in counties housing casinos.

TOPIC THIRTEEN: REVIEW OF MATERIAL FOR FIRST EXAM

Outline of first exam.
Note that each question is ten points.
1. Study the behavioral era, and the examples.
2. Study what is a good theory, and review Groupthink readings.
3. Be able to identify the unit of analysis of ten examples. Is each individual, county, state, or nation.
4. Identify the level of measurement of each of ten indicators. Is each nominal, ordinal, or interval.
5. Study test-retest reliability, and the table examples given.
6. Study convergent-discriminant validity test, review the table giving the correlation matrix.
7. Review construct/criterion validity test, and the table example given.
8. Review all class lecture material on panel studies.
9. Review all class lecture material on survey research--sampling and types of surveys.
10. Carefully review and outline chapter 5 of the textbook.

OCTOBER 6, FIRST ESSAY EXAMINATION

TOPIC FOURTEEN: DESCRIPTIVE STATISTICS

A. CENTRAL TENDENCY- typical case
1. Mode- category with a plurality (the greatest single number of cases); nominal level
2. Median- category with the "middle" case; half of the cases are below this case, and half are above; ordinal level
3. Mean- average; add up all of the cases' scores, and divide by the number of cases; interval

B. DISPERSION- diversity, how divided or united the cases are, the form of the distribution (interval level)
1. Range- distance between the scores of the extreme cases, interval
2. Variance- average squared deviation of each case from the mean. Formula on board
3. Standard Deviation- square root of the Variance.

READINGS:

1. Privatization of Municipal Services in America's Largest Cities-
Table 1- note that the mode is Satisfied with Privatization.
Table 2- the mode for most services is Satisfied. However, for drug/alcohol treatment and employment and training, the results are bimodal (two modes), with satisfied and neutral each having 7 responses.
Table 2- The last column gives the mean of satisfaction. Cities were most satisfied with the privatization of three services--street lighting, solid waste collection, and printing services.
Table 3- in each case, the most important reason for privatization was to Reduce Costs.

2. Workplace Preparedness and the Americans with Disabilities Act-
Table 1- the mode is Disagree that city policies are adequate for handling AIDS cases.
Table 3- the mode is Agree that most citizens in the community would oppose hiring someone with AIDS.
Table 5- the mode is Disagree with the three items asking about providing reasonable accommodation for someone having AIDS. Also note that mode is Agree that the budget will have to be cut to provide help to disabled workers. Mode is Undecided on whether it is too expensive to help city workers with AIDS.
Table 5 Medians- median is disagree that workers with no symptoms should receive special accommodations; median is disagree that workers in remission should receive special accommodations; median is undecided that AIDS workers should receive special accommodations; median is undecided for both of the undue hardship questions.

TOPIC FIFTEEN: CONTINGENCY TABLES

Contingency tables can be used with nominal level measures, though we usually employ ordinal or interval level data. Contingency tables permit you to view the data in an easily interpretable and understood manner.

Percentage Difference is a measure of strength of the relationship. It ranges from a low of 0 to a high of 100. Always put the independent variable at the top of the table, and the dependent variable at the side. Then, calculate the column percentages. For ordinal and interval level indicators, compare the column percents (for the two extreme categories of the predictor) across the same category of your dependent variable. Make this comparison for the two extreme categories of your dependent variable, and take the average. If one of these comparisons is contrary to your hypothesis, make the difference a negative.

Other Measures of Association to use:
1) Lambda = use with two nominal variables; is an asymmetric measure (must specify a particular dependent variable).
2) Cramer's V = use with one nominal and one ordinal variable.
3) Kendall's tau-b = use with two ordinal variables, when table is square (having same number of categories for each variable).
4) Kendall's tau-c = use with two ordinal variables, when table is not square (has different number of categories for each variable).
5) Gamma = use with two ordinal variables, produces artificially high values.
6) Somers' d = use with two ordinal variables, is an asymmetric measure.
7) Pearson correlation coefficient (r) = use with two interval or ratio variables.

All measures range from 0 for no relationship to 1 for perfect relationship. A positive or negative sign is a function of the direction of the coding of the variables and whether your hypothesis is upheld.

Multivariate crosstabulations:
Multivariate analysis involves one dependent variable and more than one independent variable (predictor).

Controlling- multivariate tables always permit you to examine the relationship between a predictor and a dependent variable, after taking into effect the impact of a second predictor.

For example, African-Americans tend to have a lower turnout than whites. A possible control variable is socioeconomic status (SES). Perhaps African-Americans have a lower average turnout than whites because of the lower socioeconomic status of blacks, and we know that people of all races having a lower SES tend to have lower turnout compared to people of all races having a higher SES. To determine whether a lower SES level explains why African-Americans tend to have lower turnouts than whites we examine: the relationship between race and turnout, controlling for SES. Do whites and blacks of the same SES level have the same turnout level; if so, SES is more important than race in shaping turnout.

___>___________>SES ________>
RACE _____________________> TURNOUT

Three types of variables that one would control for:
1) Outside variables- a variable that has an effect on one of your predictors and on your dependent variable. Here, race is an outside variable. You would control for it to determine if SES has a direct, causal effect on turnout, or whether the race-turnout effect is spurious. If spurious, then race directly affects or causes SES and turnout, but SES does not have a direct causal effect on turnout.
2) Intervening variable- a variable that is located between a predictor and a dependent variable, and that explains why the "early" predictor is related to the dependent variable. SES is an intervening variable here, as it explains why race is related to turnout.
3) Specifying or Conditional variables- a predictor that changes the relationship between another predictor and the dependent variable. That is, the relationship has a different direction or magnitude for different categories of the specifying variable. If a race gap in turnout exists only among college grads in Mississippi but not among other educational groups, then education is the specifying variable.

READINGS

1. Mainframe and PC Computing in American Cities
Table 8- each row is a different bivariate table examining the relationship between two dichotomous variables (Central System or PC Only type of computer system, and did or did not have that problem with the computer system). For each row, problems were more associated with the Central System than with PC Only type of computer system. Note that they used Cramer's V as a measure of association, which is used with one nominal and one ordinal variable; type of computer system is nominal, and they assume whether or not problems were experience is ordinal.
Table 9- each row is a different bivariate table, examining the relationship between type of computer system and positive impacts of computers. In all except one instance, PC Only systems had a more positive impact, but the differences between computer systems were generally slight.

2. The Glass Ceiling Revisited
Table 1 examines the bivariate relationships between gender and human capital and work habits. Compared to women, men generally have more education, federal service, work longer hours, travel more, relocate more, and have fewer leaves of absence.
Table 2 examines the relationship between gender and GS level, controlling for these numerous human capital and work habits. Notice that in each instance, gender discrimination in GS level persists, even after controlling for each of these possible explanations for why women have lower GS levels than men.

TOPIC SIXTEEN: STATISTICAL INFERENCE

Statistical inference is our ability to generalize a relationship found in a sample to the entire population from which that sample was drawn. That is, can we infer population characteristics from sample data. If our statistical inference test suggests that in the population the relationship between the two variables is nonrandom, the relationship is said to be statistically significant.

For example, our 1996 Mississippi Poll sampled only 601 adult Mississippians from a population of over two million. We found a definite relationship in the sample between gender and seat belt use. 60% of women said they "always" used their seat belts, compared to only 42% of men. 9% of men said they "never" used their seat belts, compared to only 4% of women. The magnitude of this relationship between gender and seat belt use was 12%: [(60-42) + (9-4)] / 2. But can we generalize this relationship found in the sample to the entire population? Is there a relationship between gender and seat belt use in the entire population? Statistical inference is the procedure we use to determine if any relationship exists in the entire population.
In this example, the chi-squared (Pearson) is 22.9 with 3 df, and is significant at .001 level. Only 1 chance in a thousand that no relationship exists in the population.

Two tests of statistical inference:

1) Chi-squared is for nominal level variables. Hence, it does not provide information about the direction of the relationship, it simply indicates that a relationship exists in the population. Since the value of chi-squared tends to increase as sample size increases, it does not measure the strength of the association between variables.

Chi-squared = summation [ (fo - f e )squared / fe ]
For the expected frequency for each cell, multiply the column total and the row total for that cell, and divide by the table total.
Degrees of freedom equal the number of columns minus 1 multiplied by the number of rows minus 1.
Consult the Chi-squared chart on page 388 of text.
On the SPSS output, use the Pearson chi-squared, which is the most widely used form.

Warning: chi-squared should not be used if any cell has an expected value less than 1, or if more than 20% of the cells have expected values less than 5.

Example from Berman's PAR (March 1997) "Dealing with Cynical Citizens," table 3.

-- OBSERVED FREQUENCIES--

-- EXPECTED FREQUENCIES--

NUMBER OF STRATEGIES

The chi-squared computation for each cell is:
(37-24.5)²/24.5 = 156.25/24.5 = 6.4
(65-77.5)²/77.5 = 156.25/77.5 = 2.0
(36-48.5)²/48.5 = 156.25/48.5 = 3.2
(166-153.5)²/153.5 = 156.25/153.5 = 1.0
Summate these four cell results: 6.4+2+3.2+1 = 12.6

Chi-squared value is 12.6 with 1 degree of freedom. (2-1) * (2-1) = 1 df. Page 388, significant at .001 level.

2) The t-test is an interval statistic (dependent variable must be interval). It tests the hypothesis that two groups have different means, and that the inter-group difference can be generalized to the population.

Two-sample t-test (SPSS-independent sample) means that each group is considered a sample.

A one-tailed t-test means that your hypothesis has a direction for the relationship. A two-tailed t-test is used to test nondirectional hypotheses. A two-tailed test is stricter, and SPSS does not report a one-tailed test, hence if your results are significant for the 2-tailed test, they will also be significant for the 1-tailed test.

Two statistics are reported-- for two populations having equal variances, or unequal variances.

The t-test is computed using the formula on page 388 of your text. Degrees of freedom equals the sum of the two sample sizes minus two. Use chart of page 389--

t-value must be larger than table entry to be significant at the specified level.

Using SPSS program. Use Compare Means- Independent Samples Statistics Menu. Your Test Variable is your dependent variable, which should be interval level. Your Grouping Variable should be a dichotomous independent variable (recode it, when necessary). Use Levine test, which must be p <= .05 for equal variances; otherwise, use unequal variances row. Cite t-value and 2-tail sig. in papers. Significance Level must be <= .05.

TOPIC SEVENTEEN: COMPUTER LAB WORK ON RESEARCH PAPER

No class held as I am presenting a paper at the Southern Political Science Association meeting.

Do the computer work to execute the research design for your research paper. Complete that work within the next two weeks. If you have any problems, contact me. You may also attend my Tuesday (Nov. 2) evening class in BE 104 to obtain computer assistance.

Also, review the course material for the second exam. If you have any questions, contact me before the test. You may also attend my Tuesday (Nov. 2) evening class in BE 104 for a brief review on some of the statistical techniques.

SECOND ESSAY EXAMINATION

TOPIC EIGHTEEN: REGRESSION- BIVARIATE AND MULTIPLE REGRESSION

BIVARIATE REGRESSION

This technique finds the best fitting straight line through a set of points. Best fitting is defined by minimizing the sum of squared distances between the points and the regression line.

Equation of line is Y = a + (b * X), where Y is dependent variable, x is independent variable, a is the Y intercept, and b is the slope of the line, or (change in Y)/(change in X)

R² is explained variance, the variance in Y explained by the independent variable's regression line.

R² = (total variation - unexplained variation)/ Total Variation

Total Variation = sum of squared distances between the mean of Y and each case's Y value

Unexplained Variation (Residual) = sum of squared distances between each case's Y value and each case's predicted Y value (from the regression equation)

Explained Variation = sum of squared distances between each case's predicted Y value and the mean of Y.

b = unstandardized regression coefficient = slope = (change in Y) / (change in X)

Beta = standardized regression coefficient = b * (sd_x/sd_y), where sd means standard deviation. It adjusts for the differing ranges and scales of the variables.

Beta ranges from -1 to +1 with 0 being no relationship between the independent and dependent variables. The sign depends on the direction of the coding of your variables. A +1 or -1 is a perfect relationship. b values have a greater range which is not confined to 1 or -1.

Pearson R is the correlation coefficient. It equals the Beta in the bivariate case only. See p. 466 of text for formula used in calculating R.

R² is the explained variation. It is the predictive ability of your independent variable.

Adjusted R² shrinks the value of R² by penalizing for each additional independent variable, and is statistically preferable to the R². See p. 440 of text.

The F statistic tests the statistical significance of the regression equation as a whole, and must be below .05. See p. 443 of text.

Problem of outlying cases. See faculty example.

MULTIPLE REGRESSION

Linear regression applied to more than one independent variable. With two independent variables, the predicted values comprise a plane (instead of a line in the one independent variable case).

Y = a + b₁x₁ + b₂x₂ + b₃x₃ + b₄x₄

b value is the unstandardized regression coefficient, controlling for the effects of all other predictors. It is used to predict the value of the dependent variable from the known values of the independent variables.

b value is also used in making comparisons across subsamples. For example, if an independent variable is more important in affecting the dependent variable among men or among women.

Beta is the standardized regression coefficient, controlling for the effects of all other predictors. It tells the relative importance of the independent variables in influencing the dependent variable. It ranges from 0 to 1, with 1 being most important and 0 being least important. Negative signs reflect the direction of variable's coding.

Multiple r is the correlation between the actual Y value and the predicted Y value from the multiple regression equation.

R² is the variance in the dependent variable explained by all of the independent variables.

TOPIC NINETEEN: REGRESSION ASSUMPTIONS, NON-LINEAR TECHNIQUES, LOGISTIC REGRESSION

REGRESSION ASSUMPTIONS

This section is drawn from William D. Berry's Understanding Regression Assumptions, Sage Publications, 1993.

Our regression model may be represented as:
Y = a + b1x1 + b2x2 + b3x3 + e,
where Y is the dependent variable, xi are the independent variables, bi are the unstandardized regression coefficients, and e is the error term.

The error term (e) represents the combined impacts of all variables influencing the dependent variable that are not included in the regression, plus any random components in the behavior of the dependent variable.

Basic assumptions of regression:

1) The effects of the independent variables are additive, meaning that each independent variable's effect on the dependent variable does not vary depending on the values of the other independent variables. In a nonadditive model, two or more variables interact in influencing the dependent variable, as shown by the effect of one independent variable varying with the value of the other independent variable (a conditional variable-effect exists).

2) Linearity assumption-- when all other independent variables are held constant, the change in the expected value of Y (Y's predicted value) associated with a fixed increase in Xi is the same regardless of the value of Xi. That is, the slope of the relationship between Xi and the expected value of Y is constant.

Excluding a squared term or an interactive term that is correlated with an included predictor will result in a biased estimator for the included variable.

There are 7 Gauss-Markov assumptions that must be met for your OLS (ordinary least squares regression) coefficient estimators (b values) to be BLUE-- the best linear unbiased estimators. Best means having the smallest sampling variance among a set of unbiased estimators; this enhances your ability to generalize to the population. Unbiased means that the estimator's mean value over an infinite number of repeated random samples is equal to the parameter being estimated. There is also an 8th assumption, needed to generalize to the population.

1) Assumption that all independent variables have nonzero variance. If a predictor has no variance, then it is unable to explain the variation in the dependent variable.

2) Assumption of absence of perfect multicollinearity.

Multicollinearity occurs if the correlation between two independent variables or among a set of independent variables is too high. Then, the standard errors for the partial slope coefficients (b values in a multiple regression equation) will be quite high, so the estimates of the effects of the independent variables will fluctuate considerably from sample to sample. Hence, one cannot obtain precise estimates of the unique effects of the independent variables.

Perfect multicollinearity occurs when the number of observations is smaller than the number of variables. Perfect multicollinearity also occurs during mistakes in handling dummy variables. If you have four dummy variables, one for each of the four regions, you have multicollinearity; you must omit one of the regions, and the Y intercept would give the effect of the omitted region. Multicollinearity also occurs when two predictors are highly interrelated.

You can detect multicollinearity by examining the correlation matrix of your independent variables. Be wary of correlations exceeding .8. Also, regress each independent variable on all of the other predictors, and be wary of multiple R's above .8.

Solutions: eliminate one of the highly intercorrelated predictors; form a scale combining the two highly intercorrelated predictors.

3) Assumption that the error term is uncorrelated with each of the independent variables. Must use theory to determine if assumption is violated, since regression residuals statistically will always be uncorrelated with the predictors.

Assumption is violated in cases of reciprocal causation, when the dependent variable influences one or more of the independent variables. Must use two-stage least squares in cases of reciprocal causation.

If a relevant explanatory variable is excluded from your regression model, and if the excluded predictor is correlated with an included predictor, then the estimator for the included variable is positively biased. If excluded variables are weakly correlated with included predictors, substantial bias will not result.

4) Assumption that the mean of the error term is zero.

If error is biased by a constant, then the intercept is inaccurate; the constant should be added to the intercept. An excluded variable that is uncorrelated with the included predictors produces a constant but nonzero error.

More seriously is if the mean of the error varies across observations, which often means that a relevant variable has been excluded from the regression equation and it is correlated with an included predictor, thereby biasing the partial slope coefficient estimators.

A varying error mean also occurs when dealing with a truncated sample-- a sample that excludes cases with lower scores or cases with higher scores on a dependent variable. It makes the independent variable estimator biased.

5) All independent variables are quantitative or dichotomous, and the dependent variable is quantitative, continuous, and unbounded. Also, all variables are measured without error.

Variables must be measured at the interval level. If ordinal data are used, you should have equal distances (in theoretical terms) between the categories. Continuousness requirement is partly met with six or more categories.

Dichotomous dependent variables require a non-linear technique such as logit or probit, rather than regression. Dichotomous dependent variables violate the assumption of a normally distributed error term, and the homoscedasticity assumption. You can receive nonsensical predictions less than 0 or greater than 1 (assuming dependent variable is coded as 0 and 1). Linearity assumption is often violated, as the actual Y values flatten into a curve when nearing 0 and 1.

Random measurement error (such as respondent guessing) in the dependent variable causes less efficient estimators and lower R2 values, though estimators remain unbiased. Random measurement error in the independent variable causes biased parameter estimates.

Nonrandom measurement error always leads to biased estimators. Examples of error: nonlinear error; error towards moderation of responses; error due to categorization, which becomes more severe as the number of categories decreases.

Measurement error is also caused by using proxy variables, since the indicator is not a true measurement of your concept.

6) Assumption of absence of autocorrelation or serial correlation (error terms for any two observations are uncorrelated). Autocorrelation causes unbiased but not BLUE estimators, and GLS (generalized least squares) should be used. Autocorrelation also causes biased estimates of standard deviation of estimators, hence statistical inference tests cannot be performed.

Autocorrelation is especially a problem in time series models, where cases are time points. Excluded predictors of the dependent variable are typically correlated over time, hence the error terms will be correlated.

Autocorrelation is seldom a problem with a random sample of a nation's population. Problem of spatial autocorrelation can arise when the units are political jurisdictions. Snowball sampling method can also cause autocorrelation.

7) Assumption of homoscedasticity, or constant error variance across categories of the predictor. Heteroscedasticity occurs if assumption violated. Heteroscedasticity causes unbiased but not BLUE estimators, and GLS (generalized least squares) should be used. It also causes biased estimates of standard deviation of estimators, hence statistical inference tests cannot be performed.

Heteroscedasticity may occur in cross-national studies, since measurement error may be greater in less developed countries, hence the error variance in such countries is greater than in developed countries.

Detect non-constant error variance across predictors' scores (heteroscedasticity) by examining scatterplot of residual plotted against each predictor. If variation is non-constant, model respecification may be necessary, since a relevant variable may have been excluded from the regression.

Example of Microcase State Welfare dataset. Region and welfare spending show unequal variance in residuals for south and non-south. Include presidential vote and tax capacity, which are relevant excluded variables.

8) Assumption that the error term is normally distributed. Relevant assumption for statistical inference tests. Normal distribution needed for small samples so that the sampling distribution of the coefficient estimators are normally distributed.

Not a problem with a large sample, since the Central Limit Theorem ensures that the distributions of regression coefficient estimators are normally distributed, even when the equation's error term is not.

CAUSAL MODELING

Multiple regression provides only the direct effects that independent variables exert on dependent variables. Yet outside variables may also affect the dependent variable by affecting an intervening variable in the model. Hence, an outside variable may exert an indirect effect on the dependent variable.

Total effects of an independent variable are equal to the sum of the direct effect of that variable and all of its indirect effects. Each indirect effect is the product of the effect that that outside variable has on an intervening variable, and the effect that the intervening variable has on the dependent variable.

Causal Modeling procedures.
1) Devise a model that shows temporal-causal ordering of the variables
2) Use multiple regression SPSS program and regress each dependent variable in the model on all of the independent variables that are "earlier" than it is
3) Draw arrows for all statistically significant linkages. Put Betas just above each line.
4) Indirect effects involve multiplying the relevant Betas together
5) Total effect = direct effect + indirect effects

TOPIC TWENTY: EXPERIMENTAL DESIGNS

EXPERIMENTAL DESIGNS

Classical Experimental Design:

Pre-test ---------------> Stimulus ----------------> Post-test
Experimental Group

Pre-test ---------------------------------------------> Post-test
Control Group

Also, both groups must be equal in composition. Ensure equality by: matching; random assignment.

Internal invalidity problems— inferences (conclusions) drawn are not an accurate reflection of what actually happened.
1) History- it affects both groups, so control needed.
2) Maturation- affects both groups, so need control.
3) Testing- pre-test in both groups, so need control.
4) Instrumentation- if post-test different from pre-test, need control group.
5) Statistical regression to the mean- if dealing with extreme cases, need control group also having extremes.
6) Selection biases- groups must be comparable.
7) Experimental mortality- check for biased mortality
8) Diffusion or Imitation- don't permit communication between the groups
9) Compensatory Rivalry- underdog effect; don't tell subjects what group they are in
10) Demoralization- don't tell subjects which group
11) Compensation- control is no longer the control

External Invalidity Problems- unable to generalize to a population
1) Sample bias- using college students, example
2) Artificial experiment- maximize realism
3) Long-lasting effect- multiple post-tests
4) Pre-test-stimulus interaction- use Solomon 4 Group design

Solomon 4 Group Design- use same two groups from the classical experimental design, include two more groups. One having stimulus-posttest only, and another having only the posttest. Must have equal groups, which then assumes equal pre-test scores.

Post-Test Only Design- one experimental and one control group, no pre-tests. Groups must be equal, use randomization

Factorial Designs- used with 2 or more stimuli

Classical Experimental Design is strong on internal validity, but weak on external validity

TOPIC TWENTY-ONE: QUASI-EXPERIMENTAL DESIGNS

QUASI-EXPERIMENTAL DESIGN

Quasi-experimental designs are only moderate on internal validity, since they are natural-occurring experiments, and people cannot be randomly assigned to the groups.

Two major types of quasi-experiments:

1) Time Series Design- multiple pre-tests before stimulus; multiple post-tests after stimulus; no control group. Failure to control for numerous threats to internal validity of quasi-experiment.

2) Control Series Design- two time series, one for experimental group, one for control group. Must have groups that are as comparable as possible. Controls for many internal validity problems.

Correlational Design— extensive social science research, such as survey research.
This is a post-test only design, with statistical controls used to simulate experimental and control group. However, random assignment is not used to create groups.

One shot case study is a pre-experiment weak on both internal and external validity. It consists of a stimulus and a post-test.

TOPIC TWENTY-TWO: ETHICAL CONCERNS

ETHICS

Stanley Milgram study- obedience to authority.

Informed Consent, components of-
1) Competence- guardians approval needed
2) Voluntarism- no deceit or over-reaching
3) Full Information- inform R of benefits and risks, can refuse answering any question, can withdraw at any time
4) Comprehension- "knowing" consent needed, don't use jargon

Anonymity versus Confidentiality-
Anonymity- no one can identify a person with their responses
Confidentiality- researcher knows who the respondent is, but promises not to tell anyone

Examples of informed consent:
1) Mississippi Poll
2) NSF Grassroots Party Activists cover letter

One must never harm subjects.
MSU Human Subjects form approval
Subpoena problem, so if confidential data convert into anonymous data as soon as possible

Studies having ethics problems:
1) MSU literacy study, when suspected interviewer fraud results in Attorney General request for R info
2) Ray Cleere's workplace study included identifiable questions and political questions, and MSU dropped out of it
3) NSF Grassroots Party Activists study- ICPSR deleted county and state variables

Political biases are a major problem in funded research:
1) Media sensationalism- 1982 Clarion-Ledger Senate poll
2) Official suppression of studies they disagree with— Mabus governmental child care study suppressed by Fordice administration

ASPA Code of Ethics: 5 sources of ethics
1) Public Interest: oppose discrimination and harassment, promote affirmative action; public right to know; involve citizens in decisionmaking
2) Legal Interest: change obsolete, counterproductive laws; prevent mismanagement of public funds, need audits; protect privileged information; whistleblower protect
3) Personal Interest: give others credit for their work-plagiarism; avoid appearance of conflict-of-interest, such as nepotism, gift acceptance, misusing public resources, improper outside employment; act nonpartisan in actions; admit own errors
4) Organizational Interests: promote creativity, open communication among workers; permit dissent, no reprisal, due process used; merit use
5) Professional Interests: keep current on new issues, problems, upgrade professional competence; professional associations active; help public service students, like internships provide

TOPIC TWENTY-THREE: UNOBTRUSIVE MEASURES AND CONTENT ANALYSIS

UNOBTRUSIVE MEASURES Problems with obtrusive measures: (see book Unobtrusive Measures by Webb, Campbell, Schwartz, and Sechrest)
1) Guinea Pig or Testing Effect: subjects may feel must leave a good impression, or test may make them interested in subject
2) Role Selection: nonrepresentative role selected, especially by less educated and less familiar with subject of test
3) Response Sets, such as acquiescence bias, sequence, wording: Mississippi Poll, spending items
4) Interviewer Effect: race, age, and sex of interviewer may affect responses

Unobtrusive Measures directly remove the researcher from the research setting.

Types of Unobtrusive Measures:
1) Physical traces- erosion and accretion

2) Simple observation- physical signs; body language; physical location (proximity to others)
Validity problems with simple observation method: available only in public settings; observations may not represent a wider population; doesn't generate explanations.

3) Archival Records- actuarial records, political records, government documents, mass media archive

4) Content Analysis- any technique for making inferences by systematically and objectively identifying specified characteristics of messages. Examples of M. Hermann's, Edith Efron's The News Twisters, and Richard Hofstetter's studies

TOPIC TWENTY-FOUR: REVIEW COURSE MATERIAL FOR FINAL EXAM

Outline of Final Examination (90 point test):
1. (10 points) Bivariate regression, use computer to type in data
2. (15 points) Multiple regression, use computer to analyze Miss Poll
3. (10 points) Classical Experimental design
4. (10 points) Internal and external validity problems with experiments
5. (10 points) Quasi-experimental design
6. (5 points) Ethics question
7. (5 points) Another ethics question
8. (10 points) Unobtrusive measures
9. (15 points) The readings on reserve since the last test

FINAL EXAMINATION: DECEMBER 15, 1999, 6 PM