We often think of data analysis as being objective. And in many cases, those presenting their analyses to us would prefer it that way. But, no matter how much we hope for it, data analysis is not objective. No way. No how. Data analysis is one of the seven steps in our data lifecycle.
The analysis step in the process involves using models to estimate, predict, and explore how things work. I really, really like playing with statistical models. They can improve lives, better allocate limited resources, and uncover and correct racism, sexism, and colonialism. However, just because something is a useful tool, doesn’t mean it’s automatically always fair, objective, transparent, or honest.
Even if you are halfway through a data project and your data is already collected, it’s not too late to use modelling to add an equity lens and get the answers that matter most to all your project’s stakeholders.
So how do we do it? What’s the secret to making data analysis more equitable?
Is Your Model Looking for Equity Issues?
For the project in the following video, we were trying to measure the increase (if any) in productivity. (Right away, you should employ critical thinking about who chose this key outcome and if it’s specific enough, but that’s another story). So, using productivity increase as our primary measure of success, we built a statistical model to tell us whether or not productivity increased as a result of the project.
We needed more than just the data on milk production before and after; luckily we had it. We have a nice dataset from a social survey that we can use for this analysis. We know the amount of milk harvested by each woman last year and this year and we know a lot about each woman’s life from the social survey:
- Whether or not she is married
- What her personal income is
- What her household’s income is
- How many times she has used the best practices she was taught
- How much time she spends on daily activities
- Her level of life satisfaction
- How many children she has
This information is self-reported and we’re going to keep that in mind, but for now, we’re going to operate under the assumption that if a woman says her life satisfaction has improved, it did.
Importantly, we also have this information from very similar women who did not receive our training, letting us create a control group in our model.
Now, let’s take a look at the different models we can come up with:
Applying an equity lens to project can help make your data analysis more equitable. Want to know more about making your data analysis more equitable? Come learn with us at We All Count, or talk to the experts at Datassist today.