Chapter 4 Research Design Comparison as the key to establishing causal relationships Multivariate nature of the world can make our first glances misleading Make sure that there is no other variable z that would influence the causal relationship between x and y How to test whether x even causes y Research designs strategies goal of these is to help us evaluate how well a theory holds up Experiments and observational studies are the two most popular strategies Experimental Research Designs Blood pressure lowering drug You would need to conduct an experiment to test the drug An experiment is a research design in which the researcher both CONTROLS and RANDOMLY ASSIGNS those value to the participants in the experiment These two features form complete definition of an experiment no other essential features A control the values of the independent variable that the participants receive are NOT determined either by the participants themselves or by nature we cannot compare those who already take the drug with those who don t Drug acts as the z variable Researchers decide who will take the drug and who won t We the researchers must only control the values of the independent variable but WE MUST ALSO ASSIGN THOSE VALUES TO PARTICPANTS RANDOMLY Must ensure that participants are divided into treatment and control groups Treatment group those who will receive the drug Control group won t receive drug will receive placebo Random assignment to treatment group ensures that the comparison we make between the treatment and the control group is as pure as possible and that some other cause of the dependent variable z won t pollute comparison Random assignment ensures the groups as a whole are IDENTICAL People who take aspirin have lower blood pressures than people who don t take aspirin non experimental design Experimental research designs decontaminate the comparison between the treatment and the control group of all other influences They successfully do this by initially putting all participants in the same pool and then randomly dividing them Experiments are uniquely capable of crossing the four causal hurdles o Evaluate whether there s a credible causal mechanism that links x to y o It is impossible for y to cause x Assigning x occurs in time before y is measured making it impossible for y to cause x If x is generated by randomness nothing can cause it o Whether x and y are correlated easily done in research design o Is there some z that is related to both x and y that makes the association between x and y spurious Experimental design does not eliminate the possibility of z it just equates the treatment and control groups It also applies to all possible confounding variables whether we are aware of them or not Internal validity experiments bring with them a particularly strong confidence in the causal inferences drawn from the analysis high levels of confidence in the conclusions about causality Random Assignment vs random Sampling Random assignment to treatment groups participants for an experiment are randomly assigned to one of the several possible values of x the independent variable Random sampling how researchers select people for participation in a study they are selected at random every member of the population has an equal probability of being selected Are there drawbacks to experimental research design 1 Esp in social sciences not every independent variable x is controllable and subject to experimental manipulation Ex impossible to experimentally manipulate a subjects gender voting analysis Independent variable is gender you cannot randomly assign Not every x can be controlled 2 Experiments often suffer from low degrees of external validity the degree to which we can be confident that the results of our analysis apply not only to the participants in the study but also the broader population It is absolutely not the case that experiments require a random sample of the target population Samples of convenience are very popular because it is very hard to get a truly random sample this is more or less the group of people that we could beg coerce entice or cajole to participate With convenience sampling it is hard to generalize to a broader population critical issue in social sciences Replication scientists implement the same procedures multiple times to see if relationship is consistent 3 Ethical dilemmas treatment of human participants in medical experiments Ex cancer patients assigned to randomly different treatments 4 When interpreting the results of an experiment we sometimes make mistakes of emphasis Experimental research designs often don t help to sort out which causes of the dependent variable have the largest effects and which ones have smaller effects Observational studies in two flavors Sometimes experimentation is impossible so experimentation isn t the most common research design Experimentation isn t applicable to everything What options are there when you cannot control the exposure to different values of the independent variable Take the world as it already exists and make the comparison between either individual units people countries or between an aggregated quantity that varies over time observational study Observational studies are NOT experiments although emulated Researches take reality as what it is and observe it attempting to sort out causal connections without the benefit of randomly assigned participants to treatment groups Different values of the independent variable already exist in the natural world scientists observe evaluate theoretical claims by putting them through the same 4 causal hurdles to discover whether x causes y Necessary that there be some degree of variability on the independent variable between cases as well as variation in the dependent variable Correlational studies some say that it is impossible to speak of causality in observational studies Observational studies face the same 4 causal hurdles 1 First causal hurdle is the same 2 Second causal hurdle is dicey x causes y y causes x possible 3 X and y covary If we fail to find covariation between x and y in an observational setting we should still proceed to the fourth hurdle because the possibility remains that we will find covariation between x and y when we control for some variable z 4 Random assignment to treatment groups which proves that there are no external factors that interfere with the causal relationship between x and y in experimental studies in absent from observational studies
View Full Document
Unlocking...