What does an observational cosmologist do?

Cosmology research is at the nexus between theoretical math and physics, but observational cosmology introduces the additional challenge of physically and statistically motivating each step then interpreting results. You could say, this is where it gets real. You can take every math and physics course under the Sun (ahem) but nothing prepares you for being a scientist like answering a question no one has asked before in a way no one has tried, starting from scratch then getting your results published in a peer reviewed journal. My adviser Dr. Lubin and colleague Dr. Lemaux showed me the importance of collaboration and mentorship in this process. Being an observational cosmologist is not just solving equations on a chalkboard, but teaching yourself the programming languages and statistics you need to get answers from the data with your own two hands (I call this nerd street smarts). So you need to know what galaxies look like? Learn to use a telescope. So you need to extract meaning from the data? Make a model that tests your hypothesis. So you think you discovered something big? Learn how to make plots to visualize the data and get your point across. Then, go back to the drawing board if the question you asked cannot be answered with the confidence to warrant a discovery. Some statistical questions we encounter along the way:

• Classifying galaxies as part of the red sequence (a line) or blue cloud (a blob), based on empirical models in color-magnitude space of observed properties.
• Fitting emission and absorption lines in spectra using multiple Gaussians versus a simple bandpass or using theoretical corrections.
• Identifying whether or not two bivariate samples are drawn from the same distributions (if they look different by-eye, how do we show this is statistically significant).
• Combining multiple data sets taken under different conditions with different selection criteria. (How do we determine which or our data is useable? How do we use data that may be less reliable so that we do not lose valuable information?)
• Completeness, selection and sampling bias. (Is a trend in the data real, or a result of our selection methods? Are all the objects in our sample representative of the underlying population? How do we determine this without infinite data? How to we account for which are representative?)
• Identifying false negatives, false positives, small number statistics, noisy data.
• Clustering in a 3D space based on two spatial dimensions and a third proxied by redshift (resampling to determine centroids and boundaries with limited data).