Two More Critiques of Grit Analyzed

<a href="http://www.afterecon.com/economics-and-finance/grit-research-is-alive-and-well/">I recently argued the grit literature remains a viable project. I primarily addressed <a href="http://psycnet.apa.org/index.cfm?fa=search.displayrecord&uid=2016-06824-001">"True Grit..." by Remfeld et al. <a href="https://twitter.com/JohnVandivier/status/834968059282665473">An informed source recommended that I consider two additional studies critical of grit:

1. Genetically-Mediated Associations Between Measures of Childhood Character and Academic Achievement (June 2016, 3 citations)
2. Much Ado About Grit: A Meta-Analytic Synthesis of the Grit Literature (June 2016, 17 citations)

Both lower in cites/time compared to the previously analyzed Remfeld et al study (15 cites, November 2016). Frankly, I think the second study and True Grit bear essentially the same message with the latter doing so in what I considered to be a more well-formed way.

Study 1 results:

1. Table 2 shows huge cross-correlation between Grit and Need for Cognition, supporting the Cold-Passion decomposition of grit.
   1. Consider that need for cognition is essentially the same as \"passion for learning.\"
   2. Consider that in the case of academic achievement, \"passion\" would be \"passion for academic achievement\" == \"passion for learning,\" so the correlation is structurally expected.
   3. However, we expect the parameterization in this study exactly ignores theorized \"cold grit\" vs passion variation because the kind of passion is coincident instead of orthogonal.
   4. To identify the effect described above there would need to be multiple kinds of passion surveyed, so we could identify an individual-level state where the person has a common factor of work ethic (cold grit) which does not vary by kind of task, but they also have passion for some particular kind of task. Eg measure need for music instead of need for cognition.
   5. In sum, the study may support the cold-passion decomposition and possible presence of a masking effect. If the effects were well separated we shouldn't expect much cross-correlation.
2. Table 3 shows large correlation inside the Big 5.
   1. If conscientiousness and agreeableness have a correlation of about .5 what's wrong with distinguishing grit which has a correlation of about .55 in this paper?
   2. Remfeld measures the grit-conscientiousness correlation a bit less at about .53, but grit improves model power to a greater extent than agreeableness.
3. Table 6 gives good evidence for the existence of a common factor among the factors analyzed.
   1. Note that this does not refute the decomposition theory of grit as the effect is expected to be masked by their method of measurement, as discussed under point 1.
   2. Note that the subfactors had significant and large loadings in the common factor model. This means the common factor will capture the trend, but the magnitude will be imprecise relative to the full model, and therefore any forecasting using the common factor will incur error. This error has economic cost which likely renders a reduced model economically suboptimal for applied use.
   3. Also, common factor models are theoretical anyway. You can't measure \"the common factor.\" It just means you measure one of the subfactors to capture the trend of the common factor. So why not let grit be that common factor parameterized? The papers seems to argue that need for cognition is the better tool, but as per the discussion under point 1 this is unreasonable because need for cognition is precisely the same as the theorized passion subfactor of grit in the case of academic achievement. So we minimally need two factors and a different survey to capture the theorized effects.
   4. The idea that the 1-factor model has just as good fit as the two factor model is flat wrong on their own data. p = .1485 is obviously an order of magnitude superior compared to p = .33. I understand on the statistician's view neither is significant at the arbitrary .1 or .05 alpha levels, but from the economists view it's obvious that there is a huge differences in economic value. Eg firms which make forecasts that are 2.2 times less likely to be erroneous would obtain large profits.

Study 2 results: