<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Chris Howell on Medium]]></title>
        <description><![CDATA[Stories by Chris Howell on Medium]]></description>
        <link>https://medium.com/@chris2000?source=rss-e566e5cf2d81------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*59OASY81pFggnQWW.</url>
            <title>Stories by Chris Howell on Medium</title>
            <link>https://medium.com/@chris2000?source=rss-e566e5cf2d81------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 19:01:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chris2000/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Where should we look for habitable planets?]]></title>
            <link>https://medium.com/data-science/where-should-we-look-for-habitable-planets-9d9330275d8c?source=rss-e566e5cf2d81------2</link>
            <guid isPermaLink="false">https://medium.com/p/9d9330275d8c</guid>
            <category><![CDATA[exoplanets]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <category><![CDATA[astronomy]]></category>
            <dc:creator><![CDATA[Chris Howell]]></dc:creator>
            <pubDate>Fri, 07 Feb 2020 14:56:46 GMT</pubDate>
            <atom:updated>2020-02-14T04:26:55.638Z</atom:updated>
            <content:encoded><![CDATA[<p>Like many people, I have been fascinated with astronomy and space exploration since I was a child. One of the most intriguing possibilities is worlds around other stars, and inhabitable (or inhabited!) worlds even more so. Before the discovery of any exoplanets, we could conjecture about the nature of other star systems but we had only our solar system to observe and check our theories. It was only natural that we assumed other star systems with potentially habitable planets would look like ours — large gas giants in larger orbits with smaller rocky planets closer to the star. However, this has not turned out to be the case.</p><p>With the discovery of 51 Pegasi b, we learned that there are planets around other sun-like stars and that these star systems are arranged differently than ours. Further exoplanet discoveries have reinforced this, including some star systems with potentially habitable worlds. One such example is Kepler-69 (shown below), which has a large gaseous planet orbiting close to the star, as well as a smaller, potentially habitable planet in a larger orbit — an arrangement very different from the solar system’s</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EKghIPqQ0-iwvyuCa6DHSQ.jpeg" /></figure><p>How do we sort through the data to find stars that are good candidates for habitable worlds? We already know that habitable planets require main sequence stars that are not too hot or big, but even with those requirements, many stars fit the bill. With limited resources to search for exoplanets, any way to narrow the search could be helpful. Perhaps we can move beyond the stellar requirements and predict where habitable planets might be using what we know about already discovered star systems. One approach might be to use machine learning to attempt to better predict which stars will have habitable planets.</p><p>With this in mind, I built a gradient boosted model to predict if a star system has a habitable planet based on the star and its other planets. As input, I used the mass, radius, and orbital size of the non-habitable planets for each star, as well as the star’s temperature, radius, and mass. Only about 4% of multi-planet star systems have habitable planets, so the baseline of guessing no star has habitable planets, while unhelpful, does present a challenge to overcome. Of course, the goal is to improve on stellar data, so I also built a similar model using only that data. Thus we can see if the planetary data improved the predictions.</p><p>Both of my models were successful (they beat the baseline in both accuracy and ROC-AUC scores as well, with precision and recall detailed below). Also, the planetary model beat the stellar-only model, as I had hoped. As further validation, I included the solar system in the test set with Earth removed (like the potentially habitable planets of the other systems) to see if the models could correctly predict the sun has a habitable planet from the sun and its other 7 planets. Happily, our sun is predicted to have a habitable planet by the planetary model! Unfortunately, the stellar-only model gives our sun just a 40% chance of having a habitable star.</p><h3>Evaluating the models</h3><p>The stellar model had a recall of 33%, correctly finding one-third of our target systems, and a precision of 50%, with an equal number of false positives and true positives. The planetary model had a recall of 50%, finding half of the habitable systems, and a precision of 75%, with only one false positive for three correct predictions. The baseline, of course, would have a recall of 0 — identifying none of the potentially habitable star systems but having no false positives either.</p><p>The gradient boost model gives us probabilities as well as classification, so we can look at the probability a star has a habitable world. This could be useful as detection methods improve or more resources are allotted to the search. The more candidates we can check, the more important the recall and the less important the precision — more false positives are an acceptable trade-off for fewer false negatives. Accepting only a 5% likelihood gets us a recall score of 100% in both models, with the planetary model again giving better precision than the stellar-only model — 24% vs 20%. As potentially habitable planets are a rare occurrence, it should not be surprising that there was a tradeoff between precision and recall, but this does show us the model is behaving as we would like — less likely predictions include more false positives but also capture all the target stars.</p><h3>Looking under the hood with Shapley values</h3><p>Plotting the Shapley values can help us see what features the model is using to make its predictions and whether or not that matches out intuitions. Below is the Shapley force plot for Wolf 1061:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NTzKwxwnJqro-4UEh-EZpQ.png" /></figure><p>And this is the Shapley force plot for our solar system</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pokNJARyTizpDElKFKpMhQ.png" /></figure><p>We can see that while the star’s features dominate the prediction, the planets’ features also contribute. In particular, our sun is especially helped by Jupiter and Venus’s orbits. This lines up broadly with what we would expect from our knowledge of stellar systems. We can see the same effect on the entire testing set with a Shapley force summary:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fI5Ea-g0_yXo-fKh3yZBnA.png" /></figure><p>Again, the stellar features dominate, but we can see that the planets are certainly having an effect as well.</p><h3>Conclusions</h3><p>The models showed that there is value in incorporating data from already discovered planets when assessing the likelihood of a star having a habitable planet. The biggest challenge to this sort of model currently is the incomplete state of the data. Exoplanets have been discovered by many teams using a variety of methods and instruments, and not all of them provide the same data for each planet discovered. This exacerbates the already small sample.</p><p>It is also worth considering that the existing sample of exoplanets is skewed by our ability to detect them. For example smaller, cooler stars can have habitable planets orbiting more closely, and closely orbiting planets are easier to detect. Thus the fact we find more habitable planets around relatively small cool stars may not reflect their true distribution.</p><h3>Data, Methods, References, Links</h3><p>Exoplanet data comes from the <a href="https://exoplanetarchive.ipac.caltech.edu/index.html">NASA Exoplanet Archive</a></p><p>Habitability data comes from the <a href="http://phl.upr.edu/projects/habitable-exoplanets-catalog/data/database">Habitable Exoplanets Catalog</a>, a project of the <a href="http://phl.upr.edu/">Planetary Habitability Laboratory at the University of Puerto Rico at Arecibo</a></p><p>Modeling was done in python with <a href="https://scikit-learn.org/stable/">scikit-learn</a> and <a href="https://xgboost.readthedocs.io/en/latest/python/python_intro.html">XGBoost</a></p><p>My python notebook can be found at my <a href="https://github.com/chowell2000/Exoplanets">GitHub repo</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9d9330275d8c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/where-should-we-look-for-habitable-planets-9d9330275d8c">Where should we look for habitable planets?</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What if more young people had voted in 2016?]]></title>
            <link>https://medium.com/data-science/what-if-more-young-people-had-voted-in-2016-7242f251f8e6?source=rss-e566e5cf2d81------2</link>
            <guid isPermaLink="false">https://medium.com/p/7242f251f8e6</guid>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[data-journalism]]></category>
            <category><![CDATA[youth-vote]]></category>
            <category><![CDATA[voting]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <dc:creator><![CDATA[Chris Howell]]></dc:creator>
            <pubDate>Fri, 10 Jan 2020 20:02:46 GMT</pubDate>
            <atom:updated>2020-01-13T17:03:14.603Z</atom:updated>
            <content:encoded><![CDATA[<h4>Could 18–24-year-olds have changed the outcome of the election if they had voted in greater numbers?</h4><p>For decades, far fewer young Americans (no matter how one defines young) have voted than their older counterparts. This leads to speculation that young people could alter the outcome of American elections just by voting in larger numbers. With that in mind, I set out to see if this would actually have been the case in 2016. Specifically, could 18–24 year olds have changed the outcome of the election if more of them had voted?</p><p>As we all know, the United States does not use the popular vote for presidential elections, so national poll results alone won’t necessarily give a good idea of how the outcome could change. The electoral college means that both geographical distribution of the young voters and their regional political preferences are relevant, and both are ignored by national data. Thus I decided to find out how vote totals would change state by state if more young people voted.</p><h4>What if all 18–24-year-olds had voted in 2016?</h4><p>Using state by state participation rates I was able to calculate the number of 18–24 year olds who did not vote in each state. Then, using exit polling data for each state, I calculated how many new votes there would be for each candidate in each state. The result showed six states would have been won by Hillary Clinton instead of Donald Trump: Arizona, Florida, Georgia, Michigan, North Carolina and Pennsylvania. This would have given Clinton 107 more electoral votes and a convincing 339–199 electoral college victory.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/828/1*42W1APZxrpt7QN43IjNBFg.jpeg" /></figure><h4>What if 18–24-year-olds had voted at the same rate as the voters 65 and older?</h4><p>Of course, expecting every 18–24 year old to vote is highly unrealistic — no demographic has turnout of even 90%, let alone 100%. Perhaps a better question to ask would be — what would have happened if 18–24-year-olds voted at the same rate as those 65 and older?</p><p>While the result is not quite as dramatic, only changing the result in two states, it is enough to change the outcome of the election. In this scenario Florida and Michigan would have been won by Hillary Clinton instead of Donald Trump. With 45 new electoral votes, Hillary would have had a 277–261 electoral college victory.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/828/1*enDNqBpXVZlysFs5oQtc0g.jpeg" /></figure><h4>Conclusion</h4><p>We can see that if every 18–24-year-old voted, Hillary Clinton would have won easily. Perhaps that should not surprise us, as no demographic participates at 100%. The more interesting conclusion is that voters 18 to 24 years old could have changed the result of the election by voting at the same rate as voters 65 and older, a much more realistic turnout.</p><p>This final plot shows the margins of victory in each of the scenarios described above, along with the actual 2016 electoral college result.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Sl_g-vLjwlnGOAHxwqeY1A.jpeg" /></figure><h4>Data sources and assumptions</h4><p>For this analysis, I assembled exit poll data, along with voter participation rates and population statistics, all broken down by age. The exit poll data was taken from Edison Research, a firm which does exit polling for a media consortium including ABC, The Associated Press, CBS, CNN, Fox and NBC. Voter participation rates and population data were taken from the US Census. Data manipulation and plotting was done using Pandas and Geopandas in python. My notebook and datasets can be found on Github <a href="https://github.com/chowell2000/Election_2016">here</a>.</p><p>Edison collected demographic voting data for 29 states, as well as national demographic voting preferences. For three states they collected voting data about 18–29-year-olds but not 18–24-year-olds. In theses states I used the 18–29-year-old data. For the remaining 21 states, I used the national 18–24-year-old voting preferences in place of state level data. However, it is worth noting that none of these states without state level 18–24-year-old exit polling was close — the margins of victory range from 12% (Delaware, won by Clinton) to 91%, with an average margin of victory of 29%. Also, none of these states changed winners when I added the additional young voters, so using national data here should not have hurt the results.</p><p>The big assumption in this analysis is that the exit polling data is accurate, as representative sampling is difficult to achieve in an exit polling situation. If those who are willing to answer an exit poll and those who don’t want to answer the poll voted differently, the results could be biased. Additionally, it is not clear what statistical methods Edison used, so it is impossible to know what assumptions they have made. Luckily, the census data is very thorough, and contains great voter participation and demographic data for all 50 states. Also, it has been shown that if voter turnout information is wrong, then it tends to overestimate voting, so if anything this analysis underestimates the potential inluence of young voters.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7242f251f8e6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/what-if-more-young-people-had-voted-in-2016-7242f251f8e6">What if more young people had voted in 2016?</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>