Research on Kiva.org data

[Pim Schaaf]

I found myself digging through this forum in a search for other people's research on data from Kiva.org. I have settled on the following list for now:

• http://adf.ly/8YlDA (skip the ad)
  

Please leave a comment if you know of more works that should be on the list.

[Pim Schaaf]

The list, being a by-product of my thesis on the effect of teams on lending behavior, now has my own thesis included. I have published large parts of my thesis on http://thesis.pimschaaf.nl/.

[Webb Phillips]

Hi Pim,

Thanks for sharing your thesis draft!

You didn't find a significant effect of Kiva team membership on lending frequency using a linear mixed model analysis, whereas Liu, et al., 2012 report a significant effect of Kiva team membership on lending frequency using OLS regression (see p. 8). Is there a reason to prefer one analysis or conclusion here over the other?

Best,

Webb

[Pim Schaaf]

Hi Webb,

Thanks for your question.

I argue to prefer one method over the other. You correctly point out that the main difference in methods is that of OLS versus LMM analysis. In my thesis I argue that LMM analysis is preferred over OLS analysis when analysing the effect of teams on lending behaviour.

Liu et al. (2012) do not explicitly recommend LMM analysis over OLS analysis. However, they do suggest more adequate methods should be tried to rule out the possibility that lenders who join teams, lend more in the first place (Liu et al., 2012 p. 510): " Even though team affiliation is positively correlated with both the lending frequency and lending amounts, we do not rule out the possibility of a selection issue, in that lenders who join teams are perhaps inclined to lend more in the first place. We are collecting additional data in ongoing work to account for this possibility."

LMM analysis as done in my study, includes assumptions in the form of a covariance structure. Through this, systematic controlling is performed for the possiblity that lenders that join teams lend more in the first place. Moreover, LMMs allow the amount of observations and the times at which observations are performed, to differ per unit of analysis (i.e. an unbalanced dataset does not harm internal validity). OLS analyses, like those done by Liu et al. (2012), do not account for these features, limiting the data that can be analysed while maintaining internal validity. Linear Mixed Model analysis does account for these features (West, 2009; Peugh & Enders, 2005). However, due to limitations of the statistical processesor used (SPSS), this advantage could not be fully exploited in my study. I would be happily employed to improve on that in further research.

Given this, I would not suggest to prefer my conclusions over those of Liu, et al. until further research, using LMM analysis, is done.

I hope this answers your question, but feel free to ask for further clarification.

Kind regards,
Pim Schaaf

For context: most of this follows that what is written on http://thesis.pimschaaf.nl/method