Deviant Cases Consistency

[Gabriel Lourenço]

Dear all,

I hope you are doing well.

A deviant case appears in a configuration of my truth table, should I delete this configuration? (see attached)

Best wishes,

Gabriel

[Adrian Dușa]

Dear Gabriel,

This is a theoretical discussion. I would first look into measurement and calibration, and make sure that your particular case really is deviant.

Then, if this is the case, and given you have plenty of cases, you might try to delete and see what the results look like.

QCA is about a discussion back and forth to the cases, with in-depth knowledge about each and every one of them, so it pretty much depends on your particular situation.

I hope this helps,

Adrian

--
You received this message because you are subscribed to the Google Groups "QCA with R" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qcawithr+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/qcawithr/c241ca4d-7efe-4917-a24a-9fd7de673409n%40googlegroups.com.

[tocfo...@gmail.com]

There are different ways to deal with this. As Adrian explained, it is a theoretical/conceptual matter that requires you to make a choice.

1) You can ignore the deviant case no matter what and take the configuration as consistent because its parameters meet the specified thresholds.

2) You can check whether the DCC is the only member of the row, which apparently is the case here. If the only member of the supposedly sufficient configuration is a non-member of the outcome, you can designate the row as not consistent (you never really "delete" a row from a truth table).

3) If there are multiple members of the row and if at least one is not a DCC, one has to decide whether the one DCC is a reason to designate the entire row as inconsistent or not. This is a matter of what weighs more for oneself: the members that are consistent (= row is sufficient), or the row members that are DCC (= configuration not sufficient).

4) One can take the traditional, original approach and try to turn the DCC into a case that is consistent or a 0,0 case, i.e. a non-member of the row and the outcome. As Adrian said, the means to achieve this is a different conceptualization, measurement or calibration of the concepts and data.

Kind regards

Ingo

[Gabriel Lourenço]

Hi everyone,

Thank you so much for clarifying this question. I am still adjusting the calibration, but I've been having doubts in this process.

My context is: I have a database about Outcome-based contracts and I am studying conditions that enable these contracts to advance specific steps. A possible problem is that I have many cases more in than out and few cases more out than in. My dependent variable is a measurement of success (advanced steps). For example, a contract concluded with good performance is a successful case, and a contract discontinued before the signature is an unsuccessful case. But I have more steps, I just exemplify the extremes.

Conditions involve characteristics of the projects and characteristics of the institutional environment. Most of the conditions I calibrated based on the distribution of variables (1° decile - fully out, median - crossover, 10° decile - fully in). I didn't find theoretical thresholds to calibrate these conditions yet.

Maybe I have more questions ahead.

Thank you,

Gabriel

[Adrian Dușa]

I just wanted to add a little on the subsequent question about the calibration thresholds.

These are normally taken from the literature, should they already exist. Ragin's classical example of calibrating the GDP is illustrative, and it does a very good job at differentiating world countries in terms of their inclusiveness in the set of developed countries.

But theoretical thresholds do not always exist, as QCA may be applied in very specific circumstances that were perhaps never researched before.

In such situations, the researcher becomes the theoretical expert and thresholds are established at those values that make a qualitative difference between the case immediately below and the one immediately above the threshold, with extensive explanations and argumentations in the methodological section of the paper.

I would definitely not use statistical measures (i.e. deciles and median) as calibration thresholds replacements. This would shift the responsibility from "theory" (or researcher's expertise) to a data driven approach that QCA is the opposite of.

My 2 cents,

Adrian