Simpson’s paradox was discussed thoroughly in the article (paper?) by Kievit et. al. that I read for class today.  Although I have been introduced to this topic before, I did not know the term for it and didn’t know how prevalent it is.  The study about the likelihood of SP in simulated data was fascinating – 1.67% of simulated cases saw the “complementary subpopulations show a sign opposite to the aggregate”.  I will absolutely be more on the lookout for Simpson’s paradox in every interpretation of data I come across.

In response to “strategies in prisoners dilemma games” , kdomjan writes, “it’s interesting that they put this under psychology, when it’s most commonly recognized as an econ term (as far as I know)”.

While prisoners dilemma is absolutely a hypothetical to show incentives, it can also be analyzed from a psychological perspective.  For example, when put into practice, why do some people not make the “optimal decision”?

The article that addresses Simpson’s Paradox begins by addressing the problem with people’s observations of data. It most clearly describes how data can derive different conclusions if the data is split into sub-groups versus all of the data being kept together. The graph that analyzes Alcohol Intake and IQ levels is extremely interesting, it is extremely evident that within the group of people, the people with higher IQ’s consume more alcohol but within each individual the less alcohol they consume the higher their IQ. This proves Simpson’s Paradox that the overall conclusion goes against what is really true. This article gives many good examples of this paradox being apparent in data.  By going on to address how it should be dealt with there are many good suggestions of how scientists should be aware of their data and how they should assess outcomes like this. I think the idea of looking at clustering within data to see if trends might be going in conflicting ways is one of the best ways to address this issue. They conclude that data has to be analyzed very carefully in order to make the correct conclusions. I thought this article did a good job of presenting examples of Simpson’s Paradox and explaining the errors is conclusions that can be made because of this. It is an interesting topic because it is probably extremely present in many data sets. Last week Jackson Hayes addressed how in data can be manipulated “I was also intrigued by the way data can easily vary based on different variables.” I think this relates to simpson’s paradox because it shows how data in many ways can represent the wrong things. All in all I think both articles bring important aspects of data analysis to light.

I thought this reading was very interesting and had several important takeaways. This article emphasizes how occurrences of the Simpson’s paradox are higher than previously thought and how important it is to consider whether a Simpson’s paradox is present in your data. This article discusses how statistical data analysis can result in biased conclusions if you fail to consider how the relationship of data may be different for the population and subgroups, which is particularly interesting given our discussions of how data is not inherently neutral. This article hammers home the point of how important it is to consider different subgroup relationships. After reading through all of the examples of how the direction/magnitude of relationships can change, I cannot help but wonder if I’ve made causal inferences about data that would not hold up if I looked at subgroup relationships.

Money, profit, and success define our world today whether we like it or not. All research and studies are conducted with a motive in mind. In Kievit’s  Simpson’s Paradox article, treatment dosage and recovery percentage is analyzed among male and female samples. The results reach a fairly contradictory conclusion, that both men and women have a negative relationship, meaning that as you increase the treatment dosage, the recovery percentage is lowered. At first glance, one might believe this to sound utterly ridiculous. But if you think about this relationship more closely, you realize that there is omitted variable bias taking place. Omitted variable bias occurs when “a variable that is correlated with both the dependent and one or more included independent variables is omitted from a regression equation.” In this example, an omitted variable is preexistent health. For example, the majority of people who need a low dosage are probably pretty healthy, so their recovery percentage is high as opposed to the people who need a very high dosage. These individuals might already have some preexisting health conditions that are very serious affecting their recovery percentage. This isn’t captured in the data, which can distort the results. Again, here, we see that data isn’t always accurate and can be easily manipulated to tell a story. For example, coffee manufacturers want to release information from studies indicating that coffee consumption is good for your health to increase consumption and profit.

Looking back at the responses for last week, I did not realize how many issues the Pew study contained.  I actually went back and read the methods.  I think all the criticisms were fair and led me to read that study with a grain of salt.  The article on Simpson’s Paradox I found especially interesting because it relates to my thesis topic.  I am writing on how (if) economic conditions affect opioid and heroin fatalities.  So far, this question has only been directly evaluated on a national level.  The goal of my thesis is to break-out the paper’s national level into more sub-groups and evaluate regions of the U.S. with variables that are more pertinent.  I don’t necessarily think the relationship will be reversed when looking at my sub-groups, but the magnitude of the relationship might change.  Reading this paper, at least, made me realize there is some merit to what I am doing.

This article was one that I did not completely understand but the more I read the more I think I got a better grasp of the topic. Our writers speak about the idea of the Simpson’s Paradox and how it is this phenomenon that implicitly connects or reverses the certain trends of individual pieces of data when they are in comparison with one another. The article addresses that it is something we should try to avoid but I am not sure as to why. Is this a connection that humans are forcing or something that happens on its own? I do not understand how we presenting information by comparing all the subgroups reveals our implicit bias. Is it bias because we choose to specifically order the data this way?

In this overview of the methodology for a study intended to show the decrease of re-incarceration rates in America’s prisons, PEW’s summary of the methods they used to carry out this study was clearly extremely biased and unprofessional. I just can’t get over the fact that a professional organization for these types of things can be so pathetic in their methodology.

I thought this article was particularly interesting and relatable to my Economic Statistics class that I took sophomore year given the complexities that arise when looking at proportions and probabilities and such. Something that echo’s a bit of what many people have said in last weeks post is the importance of being transparent with your statistics and the data you use. After reading the first example in this article, I could easily state that  a greater proportion of females are being accepted into graduate schools in comparison to males. However, after further analysis using the “combined” method, I can also conclude that a greater proportion of males are being accepted into graduate school. When somebody is presenting an unbiased pice of work, it is especially important to acknowledge the whole story and be thorough with their analysis.

Jackson Hayes

DCS104

Professor Shrout

October 4, 2018

Differences Do Not Matter Writing

Michael E. Martell’s Differences Do Not Matter: Exploring the Wage Gap article intrigued me through his argument and his ways of collecting data. Martell persuades his audience by explaining his process of how he collected his data. He compiled all previous data surrounding his topic into a “new and improved” data set. In his article, he explains the the labor discrimination and pay gap between gay and heterosexual men. Through the use of tables, Martell is able to clearly portray the unjust statistics behind being a gay man in the workforce. I was also intrigued by the way data can easily vary based on different variables. For example, Martell explains that some of his data might not even be completely accurate because of the the low labor force of same-sex behaving men in the first place. Similarly, Martell explains how he classifies someone as “same-sex behaving” can drastically change his results. For example, if he were to change one of his four classifications, his data could have held entirely different results. Overall, I found this article very informative not only about the pay gap between same-sex behaving men versus heterosexual men, but also how classifications and lack of data representation can affect the results of our data.

Welcome to WordPress. This is your first post. Edit or delete it, then start blogging!