Gerring 2017 - Qualitative Methods

Gerring, J. (2017). Qualitative methods. Annual Review of Political Science, 20, 15-36.

1min assessment
.
 * review article on qualitative methods' relationship to quantitative methods.
 * Includes: definitions, trade-offs, cross-fertilizing opportunities, some causal inference theory
 * concisely written. sounds fairly smart to me.
 * no time?
 * definitely read: conclusion, table 1 (case selection strategies)
 * Maybe read: practical case study advice (bullet points on p25), how qualitative research can help quantitative studies (p29, middle, "In this vein, ..." to p31, top), and this summary of course...

.

intro / qual and quant
division between qualitative and quantitative started at turn of 20th century; since then, qualitative work on the defensive. How the two camps often view science/knowledge production:
 * quants: science is science, no two ways about it
 * quals: knowledge depends on the person inquiring (her ontology, epistomology, etc.)
 * ...do quals emphasize a distinction between different "ways of doing science" because it justifies their existence?

Both camps agree: there are “diverse tools” as well as “shared standards”, so it is easy to either emphasize differences or commonalities. Let's look at differences: what standards should hold?

Debate fraught with highly loaded terms. Once you choose the term, it is clear on which side you are. So author goes for defining what quants and quals actually do differently, without looking at philosophy of science.

.

definitions
Polisci people have lots of associations with the terms "qual" and "quant" (like small v large-N). Author aims for a crisp definition which "resonates" with many of these common associations (but does not just equate qual/quant with another, established distinction (e.g. "idiographic" vs "nomothetic"))
 * qual: deals with non-comparable observations
 * quant: deals with comparable observations; analysing patterns of covariation "within a formal model"

Some methods of data collection are inherently qualitative because the resulting evidence is heterogeneous / made up of non-comparable observations.

Insofar as typical comparative (historical) case analysis rest solely on covariation across cases (e.g. through QCA) they are quantitative (sic). It is the within-case evidence ("mechanistic evidence" in Beach/Pederson's words) that makes these studies qualitative (according to the author's definitions) (p19)

By coding reduce qualitative information to a few comparable dimensions, and thereby making it quantitative (or amenable to quant analysis) ("The plural of anecdote is data.") (p19-20)

Choose qual methods if you are exploring the causal mechanism (not proving it), or you are interested in a single (or a few) cases (p20-21)

.

Case selection
many case-selection typologies have been proposed (p21)

his own typology (Gerring & Cojocaru 2016) on page 22:

.

.

.

The strategies to select cases (see snippet) are divided by goals: One can select cases according to these rules a) by own judgement or b) with a quantitative algorithm (!).
 * description (typical v subtype-finding)
 * causal / exploratory: finding an explanation
 * causal / diagnostic: deciding if causal effect is present (hypothesis testing)
 * causal / estimating: gauging the size of a causal effect (also hypothesis testing)

Practical advice on how to conduct case studies: (p25)
 * select relevant, unbiased sources with diverse viewpoints
 * to find new causal factor, look for those who have been neglected, are exogenous and in principle generalizable and potentially explain most variation.

.
 * take rival explanations seriously - do not discard them easily
 * generate lots of hypotheses for within-case tests
 * use counterfactual thought experiments (making only minimal counterfactual assumptions)
 * use diagrams and chronologies with lots of interrelated data

Causal inference
Qualitative research can gain if applying explicit inferential framework like set theory or Bayesian inference (which are well established in quant research), but these frameworks help little in assessing specific evidence

Case study example for level of detail (depth of evidence) required in qual research (p26-27)

Author proposes big qual research project: lots of experts each judge evidence for case-level hypotheses; combined judgement can accumulate into higher-level inferences (with confidence intervals for disagreement) (p27-28)

.

Multimethod research
qual and quant on same research question often have different goals (exploratory v diagnostic or relationship v causal mechanism)

when they have the same goal, it's triangulation. when it does not confirm they hypothesis, no problem, it's just accumulating research, and possibly a sign to differentiate the research question

nowadays fewer exclusively qual or quant studies; good because qual data serves important functions for quant studies (p29-30) quant data (even experiments!) also enriching for qual studies (p30)
 * setting up good research design (e.g. confounder identification!)
 * settling issues of attrition/compliance
 * finding the causal mechanism
 * even for generalizing: by assessing the external validity
 * avoiding bias (interesting examples p30 bottom to p31)

.

Conclusion
good summary of most things above (so you can just read the conclusion basically) (p31)

plus a distinction between research goals: The standards for (b) (e.g. replicability, cumulation) might actually hurt research for (a)!
 * a) discovery: find an explanation (an "anarchistic affair", no clear rules how to do it best)
 * b) justification: focus of methodology, clear standards

.

unclear to me
the same as "Following the axiom that where one sits determines where one stands"?
 * 1) "They may also identify with the phenomenological idea that all human endeavors, including science, are grounded in human experience.Given that experiences—which are inevitably couched in positions of differential power and status—vary, one can reasonably expect that the methods and goals of social science will also vary."

2. Evidently, in order to say anything about our our subject one needs to circumscribe it. In doing so, one defines in some phenomena and defines out other phenomena. There is no getting around the stipulative quality of definitions. However, the choices made here are non-arbitrary insofar as they resonate with everyday usage of the term (“qualitative”) and make sense of current practices.

3. introduce problems of causal identification (e.g., heterogeneity across cases that could pose a problem of noise or confounding).

4. causal / estimating: gauging the size of a causal effect

Critique
?

Discussion

 * 1) If you believe that your dependent variable (or historic process in general) is often caused by idiosyncratic features, coding and QCA is a bad choice (because in making your data quantitative, it takes away the idiosyncrasy from the data).