Progress! …And More Delays

In the days since my last blog post we have come far with our analysis. Reaching new findings can be difficult, especially since running statistical analysis is not a straightforward process here. As I’ve said in the past, Peru is about being ready for surprises and creative solutions. The largest hurdle for performing statistical analysis here is being able to access software since the licensing fees are extortionate, especially for a nonprofit. So on this occasion I used skills from my computer science minor to get around this hurdle. Because the process is a little boring and technical, I’ll leave it at this: I found a way to upload my excel files to my W&M account and then access a W&M desktop through the web to use Stata. Basically that’s the main reason I maintain that the Internet is a form of witchcraft.

Another intern and I spent a good deal of time joining data from multiple spreadsheets in a single location where we could analyze all of the information at once. Once all the relevant variables were in the same dataset, we analyzed a large number of different relationships. It’s best to go into statistical analysis with the question of “what if” rather than a full-on expectation of what you’ll find. Nevertheless we were astonished to find that the vast majority of the variables that we expected to be statistically significant in explaining the amount of income women obtained from working with the nonprofit couldn’t be shown to have any definitive effects whatsoever.

For example, the number of children a woman has does not seem to be statistically significant in explaining the amount of money she receives from weaving textiles. We had naively gone into our analysis with the assumption that a woman with six children would not have nearly as much time to weave as a woman with only one child due to the demands of the children. Alternatively we considered the possibility that a woman with more children would a) have older children to help care for the smaller children and perform basic house chores and b) spend more time weaving because of the financial needs of her children. Neither was the case. Similarly we expected that the ability of a woman to speak Spanish (the native language of the high communities of the Andes is Quechua) might be related to a woman’s income because of her ability to sell her textiles to the wave of new tourists that have flooded the region. We also suspected Spanish ability might be related to total wealth (distinct from income in that wealth is a measure of accumulated assets whereas income is a measure of the flow of money over a given time) because either women who grew up wealthier were able to study Spanish more, or because women with a better understanding of Spanish spent more years of their lives able to access the markets in Ollantaytambo and Cuzco to buy and sell their goods. Again, what we expected to find was shown to be not significant even at the 10% level.

The moral of this story is, of course, that I’m casting expectations where it would be best not to do so. I would do well to remember that anything can happen and expecting a certain outcome doesn’t necessarily make it likely. No matter what information I find it still advances my research, even if only to say that no correlation exists.

I actually wrote this post over a week ago and wasn’t able to find reliable wifi when I wasn’t in a hurry until today. Huzzah!


  1. Hi Nate. I appreciate that finding unexpected results or no results is seemingly universal when doing research. The detail you put into explaining your original thinking and how that conflicts with what the data indicate is very insightful into how the methodology for your project works. Did you end up finding any significant correlations in the data?