He said he will not ask you to do math for the standard deviation and variance but you should know what they are used for and how to get them EXAM 2 33 questions Math will be simple What is Research Design Formally Specifies the procedures we will use to answer our research question What data will we use What hypothesis test is most important Time series or Cross sectional Informally Ensures that the data we examine will provide an answer to the research question Goals of Research Design Good Research Design will account for Threats to causality The Unit of interest Temporal Spatial variation of interest Comparison Apples to Apples Comparison The fundamental building block for causal research May compare values of Y across time May compare values of Y across units Your Research Design is specifying what comparisons you will be making Compare Like with Like We must compare 2 or more units to assess if we may infer that X causes Y I e Does rainfall X affect turnout Y Is the turn out greater in a precinct that received no rain compared to a precinct that received rain MUST answer the How else question Were those 2 precincts different in ways besides rainfall that might be important How do we compare The units should be identical except for the Independent variable Y of interest Maximize Comparability Randomly assign units to the values of X Statistical control Matching by selecting cases that differ only on the key X variable i e Rallying event Rally around the flag 9 11 Bush s approval ratings Internal Validity Does our Independent Variable X really cause the change we see in our Inferring Causation Dependent Variable Y Threats to Internal Validity 1 History 2 Maturation I e President approval rating always decline For every president 3 Testing Priming 4 Regression to the mean They re having a rally bad day 5 Selection Bad sample Choosing an Average Voter compared to an Average American BIG DIFFERENCE This is when you Select into the sample How can we ensure Internal Validity Internal Validity is the goal of true Experimental Designs Isolate the true Causal Mechanism We manipulate X while holding other possible causes constant and observe changes in Y The Goal is to ELIMINATE all alternative explanations Generalizing From our study External Validity The extent to which the conclusions from our study would hold for other people in other places and at other times It s ALL about Generalization Categories of Threats to External Validity 1 People 2 Places 3 Times There are many different threats to External Validity these are just the major ones How can we ensure External Validity Use random selection of your sample whenever possible Make sure that subjects who begin your study stick with it as much as possible Get as specific as possible in terms of describing the full context of your DATA In what ways are your data points similar In what ways are your data points different Replication Replication Replication Standard Experimental Design 1 Randomly divide subjects into 2 groups Treatment group Receive Control group Placebo 2 Present Treatment group with stimulus NOT control group 3 Measure the 2 groups afterward We can infer that any difference was caused by the treatment Key Elements of Experimental Designs Random Selection Every unit has a equal chance of being selected for experiment Random Assignment Every participant has an equal chance of being placed in treatment or control group Random selection and Random assignment ensure comparing like with like Randomization makes the 2 groups identical on average in all ways except the treatment This eliminates the threat of a spurious relationship Randomization controls for observables Observables things we could measure and statistically control for Randomization controls for the unobservable Unobservable Differences between units we cannot see 3 Types of Experiments 1 Laboratory Experiments 2 Field Experiments 3 Natural Experiments Laboratory Experiments Select subjects randomly Control absolutely everything Change only one variable Physically bring subject into a lab for maximum control In political science there are 2 main types Economic experiments Monetary incentives Political psychology experiments Wide variety but no money Field Experiments Attempt to control as much as possible Expose some subjects to treatment and others not Randomly assign individuals into groups but perform the manipulation in the real world Observe difference Random assignment to receive treatment but actually receiving the treatment might NOT be truly random Natural Experiments An outside event separates control and treatment group We observe behavior Something we not cause occurs Natural disaster Terrorist attack Some other significant event Observe behavior after the change occurred Not really an experiment but we treat it as if randomization occurred ASIDE Not sure if he ll bring it up but it s in the notes A 4th type Laboratory experiments in the field Do this when interested in a laboratory experiment with a very specific population You bring the lab to them Drawbacks of Experimental Design Not all Independent Variables Y can be randomly assigned Sex Religion Age May lack external validity The lab vs the real world Selection is not always random Convenience samples The college sophomore problem testing college kids is not a sample of the real world Types of Studies 1 Theoretical Studies No data 2 Experimental Studies Data is the result of manipulation 3 Observational Studies Take the world as it is and study naturally occurring differences between units 2 types of observational studies Time series Holding a unit constant Single unit sample over many periods of time I e Obama and Presidential approval ratings over 4 years Cross sectional Holding time constant Many units samples at a single point in time I e 10 different nation s GDP for 2008 Differences between the 2 Larger temporal dimension Time series May not need to collect data on other variables However has our unit changed over time I e would Obama become a Republican Potential Problems with Time Series Multiple events happening at the same time We can t go back and time and get better data I e going back to JFK s presidency to get Sometimes small samples problems with statistics a missed approval rating for the month of march We DON T have the Counter Factual state of the world if things were different Cross sectional We need to collect data on other variables Larger Spatial dimensions And we
View Full Document
Unlocking...