Comm106 Week 3 notes
Comm106 Week 3 notes Comm106
Popular in Communication Research Methods
Popular in Communication Studies
verified elite notetaker
This 3 page Class Notes was uploaded by Erica Evans on Sunday January 24, 2016. The Class Notes belongs to Comm106 at Stanford University taught by Jennifer Pan in Fall 2016. Since its upload, it has received 16 views. For similar materials see Communication Research Methods in Communication Studies at Stanford University.
Reviews for Comm106 Week 3 notes
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 01/24/16
Comm 106 Class 5 1202016 Causality Examples of causal claims Free media causes democracy Obama won because he ran the best campaign A causal relationship can explain what has happened in the past and predict what will happen in the future I want to know the causal effect of using social media on procrastinationquot Pitfalls Association is not causation X and Y may demonstrate a simple correlation while a third variable Z is the real cause Examples Ice cream sales cause drowningquot actually in sunny weather you go swimming which leads to more downing not icecream Hair color causes party IDquot actually the causes are race and age In order for X to cause Y X must come before it But X coming before Y is not enough to prove it causes Y The fact that there is no association does not mean there is no causation The reason is there is a selection mechanism people behave differently in different situations Example Breast cancer patients that pursue more or less invasive treatment options have the same survival rates But this is because those that have less invasive treatments have cancer that is less aggressive and those that pursue more invasive treatments have more severe cases of cancer Be clear about counterfactuals Think through what that counterfactual would be The difference between what actually happened and what would have happened if x had been different in some way Example The causal effect of last night s tuna sandwich on my stomach today is the difference between how I feel now and how I would have felt had I not eaten the sandwich The problem is we can t go back in time This is the Democracy causes countries to be richer The unit of analysis is the country in a given year China 2010 Y GDP per capita X Democracy 1 to 10 We observe 2010 US GDPcapita and China GDPcapita We want to see what China s GDP would have been if it was democratic in 2010 but we can t So we try to statistically test for causality We find countries similar to China that were democratic in 2010 Then it is really important to have good comparisons in order to determine causality 1 Find subjects Avoid sample selection bias Randomly assign some subjects to treatment group and some to control group Measure outcome before and after experiment Randomization in assignment is key so treatment and control groups on averagequot will be the same on all other factors Must not allow subjects to select intoquot treatment Problems Infeasibility It is impossible to randomly assign free media to countries for example Ethical costs Deliberately assigning treatments to certain people is unethical like assigning someone to smoke for 30 years to see if it causes cancer External validity If the result can be generalized For example if the subjects are all college students this may not be representative of the whole public Internal validity If the effect of X on Y can be isolated from other explanations If the treatment is not truly randomly assigned or there is noncompliance you cannot know the effect of X on Y So whatever Y you say it may not be related to X Types Takes place in a subject s natural environment not in a lab The problem is it is harder to make sure the subject receives the treatment properly It is much easier to manipulate the treatment and you can create environments that are not available in real life Has a lot of the same problems as the field experiment Nature or the government performs the assignment for us For example the Vietnam war draft based on lottery We cannot control the assignment of X we can only observe it The assignment is not random The big problem is that we can only control for factors if we know or suspect that they are involved Maybe you survey people who exhibit a particular characteristic You have to control for other factors that you think might affect Y But it is hard to control for everything or to think of everything that may be a confounding factor Types Crosssectional compare different units different people at the same time Time series compare the same unit over time Crosssection time series or a panel set of units observed over time Ex countries from 1945 until now If you want to argue that X causes Y Make sure there is an association between X and Y unless confounders hid the association Make sure there is no lurking Z Make sure that Y is not actually causing X