POSC 341: Final Exam Study Guide Inductive vs. deductive research – Define each & explain difference between them Induction: the process of drawing an inference from a set of premises and observations. The premises of an inductive argument support its conclusion but do not prove it. Deduction: A process of reasoning from a theory to specific observations. The proper application of logic guarantees the truthfulness of a proposition. In a valid deduction, it is the structure of the argument that counts. Difference: Induction differs from deduction because the premises do not guarantee the conclusion but instead lend support to it. Induction does not rely on formal proof, but rather gives us reasons for believing in the conclusion’s truthfulness. Anecdotal vs. empirical evidence – Define each & explain difference between them Anecdotal evidence has several definitions, which usually relate to how certain types of evidence cannot be used to logically conclude something. Anecdotal evidence is often used in place of clinical or scientific evidence, and may completely ignore research or harder evidence that points to an opposite conclusion. Types of anecdotal evidence include claiming non-factual information based on the experiences of a few people, stories that would seem to contradict factual information, and word of mouth recommendations. Scientific evidence is considered empirical when it can be observed by many people and all will agree as to what they observed. An example would be reading a thermometer. No matter who observes the thermometer, it still displays the same temperature. 1 : originating in or based on observation or experience 2 : relying on experience or observation alone often without due regard for system and theory 3 : capable of being verified or disproved by observation or experiment. From my understanding it can be through naturalistic observation (the in depth observation of a phenomenon in its natural setting) or Experimental (manipulating an independent variable to observe its effects on a dependant variable). Experimental evidence is much more reliable as naturalistic observations are vulnerable to researcher bias. Difference: Anecdotal evidence doesn’t really tell us much outside of that one relationship. Often times found in the media. Don’t tell us how the world works. We want sound, systematic, empirical evidence. Deterministic vs. Probabilistic relationship – Define each & explain difference between them Deterministic Relationship: A relationship that holds true in a mathematical sense according to some preconceived rule or formula. For example, A = WL describes the relationship between the Area (A), the Width (W) and the Length (L) of a rectangle. Probabilistic Relationship: A deterministic system has a single result or set of set of results given a set of input parameters, while a probabilistic system will have results that vary. Often, a probabilistic system (also called stochastic model, process, or system) is solved with the Monte-Carlo method. In this case, a computer program uses a pseudo random number generator to provide values of the attributes in the system that vary. The program provides an assessment of the uncertainty of results. Typically, a large number of runs (also called trials or iterations) are made. Summary statistics may include the value that occur most frequently (mode), the mean value, and low and high range, for instance the 10% and 90% percentile. The standard deviation and histogram of results may also be part of the summary information. There is no single standard presentation as this will depend on the application. The alternative to Monte-Carlo methods is to solve the problems using the mathematics of probability. This can be very complicated to do in some cases Experimental design Types Classical (Controlled Experiment) Simple Post-Test Design: Involves two groups and two variables, one IV and one DV. Subjects are randomly assigned to one or the other of the two groups. The experimental group given the treatment or stimulus, the control group given the placebo. DV measured for each group. Repeated-Measurement Design: Contains several pre-treatment and post-treatment measures (used for when the researcher does not know how quickly the effect of the IV should be observed or when the most reliable pre-test measurement of the DV should be taken.) Multigroup Design: More than one experimental or control group is created so that different levels of the experimental variable can be compared. This is useful if the IV can assume several values or if the researcher wants to see the possible effects of manipulating the IV in several different ways. May involve a post-test only or both a pre-test and post-test. May also include a time series component. Field Experiments: Also known as quasi experiments. These are experimental designs applied in a natural setting. There is no random assignment of participants to groups, but the investigator does try to manipulate one or more IVs. Notation of “Classic Experimental Design” Experimenter establishes two groups: experimental (receives the stimulus) and control (no experimental manipulation) Random assignment of individuals to groups (randomization) Researcher controls the administration or introduction of the experimental treatment (test factor) Researcher establishes and measures a DV, determines whether or not there has been an experimental effect The environment of the experiment is under the experimenter’s directions. Can control or exclude extraneous factors. Strengths/weaknesses of experimental research designs Quasi-experimental/Non-experimental designs Types (PAGE 149) Small N Designs (Case studies, Comparative Analysis and focus groups): researcher examines one or a few cases of a phenomenon in considerable detail, typically using several data collection methods, such as personal interviews, document analysis and observation. Cross-Sectional Design (Surveys and Aggregate Data analysis) Longitudinal Design/ Time Series (Trend Analysis, Panel Study and Intervention Analysis) Strengths/weaknesses of these research designs Strengths External Validity Avoids Hawthorne Effect Weaknesses Internal Validity: can we be sure that x is causing y Causal relationship Non-spuriousness Levels of measurement Nominal, Ordinal, Interval – Be able to identify this based on how variable is coded Ordinal: Assumes there exists a variable X with values 1,2 and 3 where 1<2<3. It is not certain if the quantity of x is equal in all places. (Bronze, Silver, Gold) Interval: Assume there exists a variable X with variable 1,2,and 3 where 1<2<3. Same distance from 1 to 2 and from 2 to 3. (temperature) Nominal: classification of observations into a set of categories that do not have direction (do not represent quantities of that variable) (Race, religion, gender, region) Discrete, Continuous – Be able to identify this based on how variable is coded Discrete: Finite # of categories/ values on the measurement scale (weight in whole pounds) Continuous: potentially infinite # of values on the measurement scale (weight in decimals) Reliability of measures – What does this mean? How can we ensure reliability? This refers to the degree of consistency of the measurement instrument A reliable measure consistently assigns the same value (#) to the same phenomenon. We want to minimize subjective indicators. Validity of measures (Note: this is distinct from validity of research designs) What does this mean? How can we ensure validity of measurement? Be able to list a few types of validity (e.g., face validity). Causality Be able to define and explain criteria (Correlation, Temp. Precedence, Non-Spuriousness, Theory) Tautology – What does this mean? How it is a threat to causality? Internal validity – Definition? Which types of design are best for ensuring it? Explain some of the threats to it The degree to which we can be sure that the independent variable caused the dependent variable within the current sample. Means that the research procedure demonstrated a true cause-and-effect relationship that was not created by spurious factors. Randomized Controlled experiments tend to have the strongest internal validity. Threats: Exclusion of spurious variable Non-random Sample History: change in the DV due to outside forces or changes in the environment (effect of instructor quality on MKT 301 performance) Maturation: change in subjects overtime that affects the DV (Effect of drug on height) Experimental Mortality: A differential loss of subjects from experimental and control groups that effects the equivalency of groups Instrument decay: change in the device used to measure DV, producing change in measurements Statistical Regression: Regression to the mean (Madden Curse/ Sports Illustrated) Testing: effect of a pre-test on the DV (Are you racist?) External validity – Definition? Which types of design are best for ensuring it? Explain some of the threats to it The extent to which the results of an experiment can be generalized across populations, times, and settings. Threats: Non-representative Sample Small time period Small cross section (10 students vs. 100 students) Non-representative case study Selection bias. Literature review – Objectives? Some examples of common weaknesses that research papers might overcome Objectives: Determine Research Question Use JSTOR, INFOTRAC, References from articles, Google Scholar, etc. to identify relevant published research Summarize research designs and findings Draw conclusions about weaknesses/ gaps in research Justify your research as an attempt to overcome these weaknesses Structure your discussion of literature thematically to support argument that “my research contributes…” Weaknesses: Limited external validity (you use different sample to test the same hypothesis) Conflicting theories and conflicting results (you test both theories simultaneously) New hypothesis from existing theory (you introduce and test new hypothesis based on logic from previously supported theory) Poor measurement (you provide better measurement) Lack of control for all relevant variables (you control for more variables than past studies) Direct & Indirect observation – Main differences between them? Strengths/weaknesses of each Look at notes page labeled #1. Survey research methods What makes a poll valid/scientific? Representative Sample: Face-to- Face, Telephone, Exit Polling Other problems with surveys (question wording, question order, etc.) Question Wording: Biased/leading questions: encourages a particular response Double-barreled questions: including more than one attitude object or stimuli in the question Ambiguous Questions: asks for evaluation of a concept that is not clearly defined Negative and double negatives: can confuse R’s and bias results Question Order: Saliency Effect (priming): Earlier questions effect later questions Redundancy: unwilling to repeat themselves Consistency: trying to appear consistent Fatigue: less consideration to later questions Response Order (Response set effect): string of questions with the same responses Interviewer Bias Social Desirability: Questions that request answers that would reveal socially undesirable trait/option Gender/Race of interviewer: R’s alter responses based on perceived characteristics of interviewer. Accessibility (Assuming political knowledge): wording of questions must be accessible to most R’s Ethical concerns – What are common threats to ethical research? Harm – Types? How does this relate to ethical research? Privacy – Types? How does this relate to ethical research? Informed consent – Types? How does this relate to ethical research? What body governs ethical research at Clemson? Institutional Review Board Dependent variable – Define? Be able to identify the DV if you’re given an example of a research design. Independent variable – Define? Be able to identify the IV if you’re given an example of a research design. Unit of analysis – Define? Be able to identify the unit of analysis if you’re given an example of a research design. The social entities whose characteristics are the focus of the study. The level at which we’re observing the phenomenon. May include individuals, groups, programs, organizations and institutions, cities, states, nations, etc. Hypotheses Null vs. research/alternative Null: states that there is no relationship between the IV and DV Research/ Alternative: There is a positive relationship between exposure to negative ads and turnout. Directional vs. non-directional Directional: applies to cases where IV and DV are orderable variables (ranked from high to low: Ordinal, Interval and Ratio) Non-Directional: applies to variables that are not orderable (Nominal) Descriptive statistics (Mean, Median, Mode, Range, Standard Deviation, Variance) What they are conceptually? How to calculate each of these? Mean: Average Frequency Distributions (Frequency, relative frequency, cumulative frequency, etc.) How to calculate and interpret Z-scores – What they are conceptually? How to calculate? (You will be given the z-table if this appears on exam) Probability – You should understand the items listed below, and be able to answer problems similar to the examples from class and from the problem set Probability Tree – How to set up and use to calculate probabilities Multiplication and Addition rules Mutually exclusive – what it means, and how formulas change based on this Independence – what it means, and how formulas change based on this Confidence intervals – What they are conceptually Type I vs. Type II errors – Definition? Which of these is worse? How to test the relationship between variables? T-test When to perform this test? What level of data (DV and IV) makes this appropriate? Dichotomous level variable Interval level variable What is model testing conceptually? Null hypothesis = the means for both groups are the same Alternate (research) hypothesis = there is a statistically significant difference between the means for the groups How to interpret results? Significance level for Mean Difference in t-test for Equality of Means This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If this is below .05, there is a significant difference (and thus a significant relationship between the variables) ANOVA & Scheffe test When to perform this test? What level of data (DV and IV) makes this appropriate? Nominal (including dichotomous) or Ordinal level variable Interval level variable What is model testing conceptually? Null hypothesis = the means for all of groups are the same Alternate (research) hypothesis = there is a statistically significant difference between the means of at least two of the groups How to interpret results? Significance level associated with F-statistic This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If this is below .05, there is a significant difference (and thus a significant relationship between the variables) Significance level associated with each Mean Difference in Scheffe table The Scheffe tells you which pairs have means that are significantly different from one another. If the sig. level for the difference is below .05, the difference is statistically significant (and thus there is a significant relationship between the variables) o Chi-square When to perform this test? What level of data (DV and IV) makes this appropriate? Nominal level variable (including dichotomous) Nominal (including dichotomous) or Ordinal level variable What is model testing conceptually? Null hypothesis = there is no significant relationship between these variables Alternate (research) hypothesis = there is a significant relationship between these variables How to interpret results? Chi-squared statistic Significance level associated with Chi-squared statistic This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If the sig. level is below .05, there is a significant relationship between these variables. If it is significant, you then need to look in the table itself and compare the expected values (“expected count”) to the observed values (“Count”). The differences between these tells you the nature/direction of the relationship o Gamma When to perform this test? What level of data (DV and IV) makes this appropriate? Ordinal level variable Ordinal level variable What is model testing conceptually? Null hypothesis = there is no significant relationship between these variables Alternate (research) hypothesis = there is a significant relationship between these variables How to interpret results? Gamma statistic If the sig. level is below .05, interpret the +/- sign on the Gamma statistic. If it is positive, that means there are more concordant pairs than discordant pairs. This means that as one variable increases, so does the other. If the Gamma statistic is negative, that means there are more discordant pairs than concordant pairs. This means that as one variable increases, the other variable decreases. To determine what this means substantively, you need to look at how the two variables are coded. What does it mean substantively for each of these ordinal variables to “increase/decrease” based on how they’re coded? Significance level associated with Gamma statistic This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If the sig. level is below .05, there is a significant relationship between these variables. o Correlation When to perform this test? What level of data (DV and IV) makes this appropriate? Interval level variable Interval level variable What is model testing conceptually? Null hypothesis = there is no significant relationship between these variables Alternate (research) hypothesis = there is a significant relationship between these variables How to interpret results? Correlation coefficient (Pearson’s r) This tells you how tightly clustered the data points are to the “best-fitting line.” If this number approaches 1 or -1, this means that the data points are scattered in a fairly linear pattern Significance level associated correlation coefficient This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If the sig. level is below .05, there is a significant relationship between these variables. Scatterplots What does a positive relationship look like? Negative relationship? A perfect relationship (Pearson’s r = 1 or –1) o Regression When to perform this test? What level of data (DV and IV) makes this appropriate? DV – Interval IV – Dichotomous, Ordinal, Interval What is model testing conceptually? Null hypothesis = there is no significant relationship between these variables Alternate (research) hypothesis = there is a significant relationship between these variables How to interpret results? Adjusted R-squared This tells you how much variation in the dependent variable is explained by the independent variable(s) Slope Coefficient for each independent variable This tells you how much, on average, the dependent variable changes given a one unit change in that independent variable. The key here is to interpret what a “oneunit change” means based on how the IV is coded. Significance level for each independent variable This sig. level tells you the probability of finding a relationship of this magnitude by random chance. If the sig. level is below .05, there is a significant relationship between these variables. Constant (intercept) This tells you the value of the dependent variable when the value of the independent variable(s) is zero