Analysis of Insight Data Science Fellows

May 10, 2018


The Insight Data Science Fellows program is a highly sought-after post-doctoral training fellowship to help PhD graduates find a good job in data science. I was curious about the background of the fellows (e.g. how many are from UC San Diego? Neuroscience?) and where they ended up. I couldn’t find any aggregated information about them, so I scraped their list of fellows and summarized them in the following figures (see Jupyter Notebook).

UPDATE: After completing the program, I wrote this post on its pros an cons.

Insight fellows dataset overview

For each fellow, I scraped 6 features:

  1. Academic field (e.g. Physics)
  2. Last academic position (PhD or Postdoc)
  3. Academic institution (e.g. UC San Diego)
  4. Title (e.g. Senior data scientist)
  5. Company (e.g. Twitter)
  6. Insight project title (e.g. “Parental Guidance: Finding Yelp reviews written by parents”)

In total, 794 fellows were scraped (considerably less than Insight’s advertised >1400 fellows). I was unable to get academic backgrounds of 68 of these fellows because of difficult in parsing their descriptions, leaving 726 fellows to analyze. Of these fellows, 65% recently finished a PhD and 29% are recovering from a postdoc. 18% of the fellows went on into senior data science positions.

Figure 1. Number of fellows at each company that has hired at least 3 fellows.
Figure 2. Number of fellows coming from each academic field.
Figure 3. Number of fellows from each university with at least 5 fellows.
Figure 4. Relative frequency of words used in the titles for the fellows' Insight projects.

Relation between fellow background and data science position

I was curious if there were any correlations between where a fellow came from and where they ended up (e.g. does a certain company tend to recruit from a certain field?). To analyze this, I created crosstabs between each pair of fellow features and analyzed whether there were any potentially significant biases. For the most part, there were very few significant correlations, but I came up with the following preliminary hypotheses. Because I tested hundreds of potential hypotheses, and these effects are generally small, claims of statistical significance are not appropriate here.

  1. Fellows going to Facebook and Gartner rarely have senior positions, while those at Insight Data Science always have senior positions, and ~50% fellows at Netflix and ~35% of fellows at LinkedIn have senior positions.
  2. Neuroscientists disproportionately work at Insight Data Science. Square disproportionately takes physicists, and Facebook samples roughyl equally from all fields.
  3. Fellows from biology are disproportionately likely to have senior data science positions (22%), while those from Math are less likely (10%)
  4. Fellows whose project names contain the word “find” are disproportionately likely to come from Harvard or Columbia and go to StitchFix or Yelp
  5. Fellows from Caltech disproportionately have senior positions (7/10 vs. 18% baseline)
  6. NYU, UC Davis, and U Penn fellows are disproportionately from neuroscience.

I noticed multiple differences between the PhD and postdoc fellow populations:

  1. Senior positions are more likely to be obtained by fellows with post docs (22%, 47/211) than fellows just with PhDs (17%, 78/472).
  2. Postdocs disproportionately analyzed Twitter in their Insight project (14/211, 7%) compared to PhD fellows (11/472, 2%)
  3. Fellows from NYU disproportionately tend to be postdocs (11/22, 50% compared to 27% baseline)
  4. Neuroscience fellows are more likely to be postdocs (43/102, 42%), while social science and engineer fellows are less likely to be postdocs (16% and 15%, respectively)