Experimental methods to control variance should guide us in how to select ways to mitigate threats to validity.
We're skipping up over the time-based designs in Babbie, but having a basic understanding of what's involved in time series is helpful. Want to see whether some change has an effect. To do that, you need two points in time. But interviews that are retrospective do not offer time comparisons.snapshots of social media are one wayhigher education: entering college freshmen in their first year, every year-->it's a cohort designthere are other types, each with its own tradeoffs
the drivers/speeding-->regression to the mean
in both cases they've gone back to archival records and showed how you can get two points in time
but whenever you deal with archival records, you have to account for how the recordkeeping can create artifacts (things that shouldn't be there) [shift changes, ambient things]
key vocabulary
independent variable: the stimulus, which you manipulate to determine effect
dependent variables: effect
T1 measure the dependent variables for baseline [pre test]
T2 introduce the stimulus
T3 measure the dependent variables again [post test]
the difficulty lies in selecting the samples so they match in both conditions
"all about isolating variance" -- always trying to do this, but most explicit in experiments
divide people into categories that are as discrete as possible
the more continuous, fuzzy, variables are the dependent variables
experience in this one is also an IV?
path model: survivors --> represent experience --> artifacts
the represent experience variable is an intervening variable; we need to know how the people define their experience
MT: census, school records, etc. would be one way to follow the movement of the population and compare with the sample to see how representative it is
Rob: how do you differentiate multiple pre and post tests from a time series experiment?
When you can't assign people to the different groups in each condition, it becomes a quasi-experiment-->Campbell
p. 241 shows two cases of your variable
the more heterogeneous (wider the variance within) each group, the more subjects you need
list in Babbie on 238
the longer the study, the more ppl you need in the study bc attrition over time
the "all's well" literature bias-->don't assume tktk
"hot stuff" bias
anticipate the criticisms, explain each choice I make, and acknowledge what I can't control
more important to be able to explain your choices, how much strength how much claim you can make internal and external validity
Mario's paper
How do survivors represent their experience in material artifacts?
gets at some sense of meaning that transcends the particular case; could be asked in multiple environments
then operationalize survivors, experience, representation, material artifacts
survivors: living in ES?; living there bt x and y dates?; descendants?
represent: what was chosen to keep and preserve
material artifacts: photos, papers, newspapers, records, audio, video, household objects
IV: could do by age, generation, gender, political role,
IV: survivors
person in ES 1979-1992
= 50%
<= 50%
first-generation descendants
second-generation descendants
could look at experience in a few ways: things associated with the deeply personal, and the things that are a more public representation
could do a pilot test study on the diaspora with ppl here in LA
Natascha's project
Does Twitter use affect student engagement in a German language classroom?
conditions: use T; not use T
how operationalize engagement?
finding something that is comparable for a control group is difficult; you need to have two forms of social interaction (say, public versus private)
neutral grading
tasks have to be as close as possible
effect has to be tk
measure of student engagement
is there another method that will help her get to the thing she wants to see?
if you want to look at the effect of a technology, you have to use this experiment method-->set a minimum and measure which ones do it more
minimum threshold and rate of posting
but then you have a problem with the base rate of social media interaction
MT: does Twitter make ppl focus on the key things more, since it is limited in length?
MT: ranking favorites?
bc the question is causal, you have to use experiment to isolate
Dustin's article analysis-->the researcher uses available data, which is a limitation; the significance of the finding is hard to discern
Borgman: can you get a control group with and without television? this is a snapshot in time;
there was a notable change in media consumption
the risk is that the groups are moving together-->historical threat to validity
since you only look at this slice (big data/little data problem), you don't see the effect; too narrow
what might you have done if you could go back? bc it's Public Opinion Quarterly, the audience is the one that does big public opinion polls; they know sampling and matching inside out and backwards (only need to show that they've done what's expected; the Schram literature was a landmark in the field; people reading this would know that material intimately; the cleverness of this piece is finding a way to get a control group with and without television (if you could find someone who hadn't watched tv: Nell); this is a shot in time when you could make that comparison; the rollout of tv created records that a lot of others could then use; surprising that there's not more discussion. the results is often fairly short, but the discussion is the meaning explanation; this was an early quasi-experimental piece; it's a rigorous design and it's a snapshot in time that lets us see how hard it is to get control groups on things; as the internet became a household item, ppl went back to the tv studies and tried to replicate; the landmark of how a new medium changed social engagement
a whole lot of research came out of this: "tv doesn't tell you what to think but it tells you what to talk about"
MT: how does that compare to what we think about social media?
this makes a set of assumptions: we assume that media use is a zero-sum game; pairwise comparison, but notice it's only comparing two (not newspaper and radio)
then moved into time-shifting media use; asynchronous media
series of shifts