Simply Statistics

25
Aug

Interview with COPSS award Winner John Storey

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

jdstorey

 

Editor's Note: We are again pleased to interview the COPSS President's award winner. The COPSS Award is one of the most prestigious in statistics, sometimes called the Nobel Prize in statistics. This year the award went to John Storey who also won the Mortimer Spiegelman award for his outstanding contribution to public health statistics.  This interview is a particular pleasure since John was my Ph.D. advisor and has been a major role model and incredibly supportive mentor for me throughout my career. He also did the whole interview in markdown and put it under version control at Github so it is fully reproducible. 

SimplyStats: Do you consider yourself to be a statistician, data scientist, machine learner, or something else?

JS: For the most part I consider myself to be a statistician, but I’m also very serious about genetics/genomics, data analysis, and computation. I was trained in statistics and genetics, primarily statistics. I was also exposed to a lot of machine learning during my training since Rob Tibshirani was my PhD advisor. However, I consider my research group to be a data science group. We have the Venn diagram reasonably well covered: experimentalists, programmers, data wranglers, and developers of theory and methods; biologists, computer scientists, and statisticians.

SimplyStats: How did you find out you had won the COPSS Presidents’ Award?

JS: I received a phone call from the chairperson of the awards committee while I was visiting the Department of Statistical Science at Duke University to give a seminar. It was during the seminar reception, and I stepped out into the hallway to take the call. It was really exciting to get the news!

SimplyStats: One of the areas where you have had a big impact is inference in massively parallel problems. How do you feel high-dimensional inference is different from more traditional statistical inference?

JS: My experience is that the most productive way to approach high-dimensional inference problems is to first think about a given problem in the scenario where the parameters of interest are random, and the joint distribution of these parameters is incorporated into the framework. In other words, I first gain an understanding of the problem in a Bayesian framework. Once this is well understood, it is sometimes possible to move in a more empirical and nonparametric direction. However, I have found that I can be most successful if my first results are in this Bayesian framework.

As an example, Theorem 1 from Storey (2003) Annals of Statistics was the first result I obtained in my work on false discovery rates. This paper first appeared as a technical report in early 2001, and the results spawned further work on a point estimation approach to false discovery rates, the local false discovery rate, q-value and its application to genomics, and a unified theoretical framework.

Besides false discovery rates, this approach has been useful in my work on the optimal discovery procedure as well as surrogate variable analysis (in particular, Desai and Storey 2012 for surrogate variable analysis).  For high-dimensional inference problems, I have also found it is important to consider whether there are any plausible underlying causal relationships among variables, even if causal inference in not the goal. For example, causal model considerations provided some key guidance in a recent paper of ours on testing for genetic associations in the presence of arbitrary population structure. I think there is a lot of insight to be gained by considering what is the appropriate approach for a high-dimensional inference problem under different causal relationships among the variables.

SimplyStats: Do you have a process when you are tackling a hard problem or working with students on a hard problem?

JS: I like to work on statistics research that is aimed at answering a specific scientific problem (usually in genomics). My process is to try to understand the why in the problem as much as the how. The path to success is often found in the former. I try first to find solutions to research problems by using simple tools and ideas. I like to get my hands dirty with real data as early as possible in the process. I like to incorporate some theory into this process, but I prefer methods that work really well in practice over those that have beautiful theory justifying them without demonstrated success on real-world applications. In terms of what I do day-to-day, listening to music is integral to my process, for both concentration and creative inspiration: typically King Crimson or some variant of metal or ambient – which Simply Statistics co-founder Jeff Leek got to endure enjoy for years during his PhD in my lab.

SimplyStats: You are the founding Director of the Center for Statistics and Machine Learning at Princeton. What parts of the new gig are you most excited about?

JS: Princeton closed its Department of Statistics in the early 1980s. Because of this, the style of statistician and machine learner we have here today is one who’s comfortable being appointed in a field outside of statistics or machine learning. Examples include myself in genomics, Kosuke Imai in political science, Jianqing Fan in finance and economics, and Barbara Engelhardt in computer science. Nevertheless, statistics and machine learning here is strong, albeit too small at the moment (which will be changing soon). This is an interesting place to start, very different from most universities.

What I’m most excited about is that we get to answer the question: “What’s the best way to build a faculty, educate undergraduates, and create a PhD program starting now, focusing on the most important problems of today?”

For those who are interested, we’ll be releasing a public version of our strategic plan within about six months. We’re trying to do something unique and forward-thinking, which will hopefully make Princeton an influential member of the statistics, machine learning, and data science communities.

SimplyStats: You are organizing the Tukey conference at Princeton (to be held September 18, details here). Do you think Tukey’s influence will affect your vision for re-building statistics at Princeton?

JS: Absolutely, Tukey has been and will be a major influence in how we re-build. He made so many important contributions, and his approach was extremely forward thinking and tied into real-world problems. I strongly encourage everyone to read Tukey’s 1962 paper titled The Future of Data Analysis. Here he’s 50 years into the future, foreseeing the rise of data science. This paper has truly amazing insights, including:

For a long time I have thought I was a statistician, interested in inferences from the particular to the general. But as I have watched mathematical statistics evolve, I have had cause to wonder and to doubt.

All in all, I have come to feel that my central interest is in data analysis, which I take to include, among other things: procedures for analyzing data, techniques for interpreting the results of such procedures, ways of planning the gathering of data to make its analysis easier, more precise or more accurate, and all the machinery and results of (mathematical) statistics which apply to analyzing data.

Data analysis is a larger and more varied field than inference, or incisive procedures, or allocation.

By and large, the great innovations in statistics have not had correspondingly great effects upon data analysis. . . . Is it not time to seek out novelty in data analysis?

In this regard, another paper that has been influential in how we are re-building is Leo Breiman’s titled Statistical Modeling: The Two Cultures. We’re building something at Princeton that includes both cultures and seamlessly blends them into a bigger picture community concerned with data-driven scientific discovery and technology development.

SimplyStats: What advice would you give young statisticians getting into the discipline now?

JS: My most general advice is don’t isolate yourself within statistics. Interact with and learn from other fields. Work on problems that are important to practitioners of science and technology development. I recommend that students should master both “traditional statistics” and at least one of the following: (1) computational and algorithmic approaches to data analysis, especially those more frequently studied in machine learning or data science; (2) a substantive scientific area where data-driven discovery is extremely important (e.g., social sciences, economics, environmental sciences, genomics, neuroscience, etc.). I also recommend that students should consider publishing in scientific journals or computer science conference proceedings, in addition to traditional statistics journals. I agree with a lot of the constructive advice and commentary given on the Simply Statistics blog, such as encouraging students to learn about reproducible research, problem-driven research, software development, improving data analyses in science, and outreach to non-statisticians. These things are very important for the future of statistics.

24
Aug

The Next National Library of Medicine Director Can Help Define the Future of Data Science

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

The main motivation for starting this blog was to share our enthusiasm about the increased importance of data and data analysis in science, industry, and society in general. Based on recent initiatives, such as BD2k, it is clear that the NIH is also enthusiastic and very much interested in supporting data science. For those that don't know, the National Institutes of Health (NIH) is the largest public funder of biomedical research in the world. This federal agency has an annual budget of about $30 billion.

The NIH has several institutes, each with its own budget and capability to guide funding decisions. Currently, the missions of most of these institutes relate to a specific disease or public health challenge.  Many of them fund research in statistics and computing because these topics are important components of achieving their specific mission. Currently, however, there is no institute directly tasked with supporting data science per se. This is about to change.

The National Library of Medicine (NLM) is one of the few NIH institutes that is not focused on a particular disease or public health challenge. Apart from the important task of maintaining an actual library, it supports, among many other initiatives, indispensable databases such as PubMed, GeneBank and GEO. After over 30 years of successful service as NLM director, Dr. Donald Lindberg stepped down this year and, as is customary, an advisory board was formed to advice the NIH on what's next for NLM. One of the main recommendations of the report is the following:

NLM  should be the intellectual and programmatic epicenter for data science at NIH and stimulate its advancement throughout biomedical research and application.

Data science features prominently throughout the report making it clear the NIH is very much interested in further supporting this field. The next director can therefore have an enormous influence in the futre of data science. So, if you love data, have administrative experience, and a vision about the future of data science as it relates to the medical and related sciences, consider this exciting opportunity.

Here is the ad.

 

 

 

21
Aug

Interview with Sherri Rose and Laura Hatfield

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

 

Sherri Rose and Laura Hatfield

Rose/Hatfield © Savannah Bergquist

Laura Hatfield and Sherri Rose are Assistant Professors specializing in biostatistics at Harvard Medical School in the Department of Health Care Policy. Laura received her PhD in Biostatistics from the University of Minnesota and Sherri completed her PhD in Biostatistics at UC Berkeley. They are developing novel statistical methods for health policy problems.

SimplyStats: Do you consider yourselves statisticians, data scientists, machine learners, or something else?

Rose: I’d definitely say a statistician. Even when I'm working on things that fall into the categories of data science or machine learning, there's underlying statistical theory guiding that process, be it for methods development or applications. Basically, there's a statistical foundation to everything I do.

Hatfield: When people ask what I do, I start by saying that I do research in health policy. Then I say I’m a statistician by training and I work with economists and physicians. People have mistaken ideas about what a statistician or professor does, so describing my context and work seems more informative. If I’m at a party, I usually wrap it up in a bow as, “I crunch numbers to study how Obamacare is working.” [laughs]

 

SimplyStats: What is the Health Policy Data Science Lab? How did you decide to start that?

Hatfield: We wanted to give our trainees a venue to promote their work and get feedback from their peers. And it helps me keep up on the cool projects Sherri and her students are working on.

Rose: This grew out of us starting to jointly mentor trainees. It's been a great way for us to make intellectual contributions to each other’s work through Lab meetings. Laura and I approach statistics from completely different frameworks, but work on related applications, so that's a unique structure for a lab.

 

SimplyStats: What kinds of problems are your groups working on these days? Are they mostly focused on health policy?

Rose: One of the fun things about working in health policy is that it is quite expansive. Statisticians can have an even bigger impact on science and public health if we take that next step: thinking about the policy implications of our research. And then, who needs to see the work in order to influence relevant policies. A couple projects I’m working on that demonstrate this breadth include a machine learning framework for risk adjustment in insurance plan payment and a new estimator for causal effects in a complex epidemiologic study of chronic disease. The first might be considered more obviously health policy, but the second will have important policy implications as well.

Hatfield: When I start an applied collaboration, I’m also thinking, “Where is the methods paper?” Most of my projects use messy observational data, so there is almost always a methods paper. For example, many studies here need to find a control group from an administrative data source. I’ve been keeping track of challenges in this process. One of our Lab students is working with me on a pathological case of a seemingly benign control group selection method gone bad. I love the creativity required in this work; my first 10 analysis ideas may turn out to be infeasible given the data, but that’s what makes this fun!

 

SimplyStats: What are some particular challenges of working with large health data?

Hatfield: When I first heard about the huge sample sizes, I was excited! Then I learned that data not collected for research purposes...

Rose: This was going to be my answer!

Hatfield: ...are very hard to use for research! In a recent project, I’ve been studying how giving people a tool to look up prices for medical services changes their health care spending. But the data set we have leaves out [painful pause] a lot of variables we’d like to use for control group selection and... a lot of the prices. But as I said, these gaps in the data are begging to be filled by new methods.

Rose: I think the fact that we have similar answers is important. I’ve repeatedly seen “big data” not have a strong signal for the research question, since they weren’t collected for that purpose. It’s easy to get excited about thousands of covariates in an electronic health record, but so much of it is noise, and then you end up with an R2 of 10%. It can be difficult enough to generate an effective prediction function, even with innovative tools, let alone try to address causal inference questions. It goes back to basics: what’s the research question and how can we translate that into a statistical problem we can answer given the limitations of the data.

SimplyStats: You both have very strong data science skills but are in academic positions. Do you have any advice for students considering the tradeoff between academia and industry?

Hatfield: I think there is more variance within academia and within industry than between the two.

Rose: Really? That’s surprising to me...

Hatfield: I had stereotypes about academic jobs, but my current job defies those.

Rose: What if a larger component of your research platform included programming tools and R packages? My immediate thought was about computing and its role in academia. Statisticians in genomics have navigated this better than some other areas. It can surely be done, but there are still challenges folding that into an academic career.

Hatfield: I think academia imposes few restrictions on what you can disseminate compared to industry, where there may be more privacy and intellectual property concerns. But I take your point that R packages do not impress most tenure and promotion committees.

Rose: You want to find a good match between how you like spending your time and what’s rewarded. Not all academic jobs are the same and not all industry jobs are alike either. I wrote a more detailed guest post on this topic for Simply Statistics.

Hatfield: I totally agree you should think about how you’d actually spend your time in any job you’re considering, rather than relying on broad ideas about industry versus academia. Do you love writing? Do you love coding? etc.

 

SimplyStats: You are both adopters of social media as a mechanism of disseminating your work and interacting with the community. What do you think of social media as a scientific communication tool? Do you find it is enhancing your careers?

Hatfield: Sherri is my social media mentor!

Rose: I think social media can be a useful tool for networking, finding and sharing neat articles and news, and putting your research out there to a broader audience. I’ve definitely received speaking invitations and started collaborations because people initially “knew me from Twitter.” It’s become a way to recruit students as well. Prospective students are more likely to “know me” from a guest post or Twitter than traditional academic products, like journal articles.

Hatfield: I’m grateful for our Lab’s new Twitter because it’s a purely academic account. My personal account has been awkwardly transitioning to include professional content; I still tweet silly things there.

Rose: My timeline might have a cat picture or two.

Hatfield: My very favorite thing about academic Twitter is discovering things I wouldn’t have even known to search for, especially packages and tricks in R. For example, that’s how I got converted to tidy data and dplyr.

Rose: I agree. I think it’s a fantastic place to become exposed to work that’s incredibly related to your own but in another field, and you wouldn’t otherwise find it preparing a typical statistics literature review.

 

SimplyStats: What would you change in the statistics community?

Rose: Mentoring. I was tremendously lucky to receive incredible mentoring as a graduate student and now as a new faculty member. Not everyone gets this, and trainees don’t know where to find guidance. I’ve actively reached out to trainees during conferences and university visits, erring on the side of offering too much unsolicited help, because I feel there’s a need for that. I also have a resources page on my website that I continue to update. I wish I had a more global solution beyond encouraging statisticians to take an active role in mentoring not just your own trainees. We shouldn’t lose good people because they didn’t get the support they needed.

Hatfield: I think we could make conferences much better! Being in the same physical space at the same time is very precious. I would like to take better advantage of that at big meetings to do work that requires face time. Talks are not an example of this. Workshops and hackathons and panels and working groups -- these all make better use of face-to-face time. And are a lot more fun!

 

20
Aug

If you ask different questions you get different answers - one more way science isn't broken it is just really hard

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

If you haven't already read the amazing piece by Christie Aschwanden on why Science isn't Broken you should do so immediately. It does an amazing job of capturing the nuance of statistics as applied to real data sets and how that can be misconstrued as science being "broken" without falling for the easy "everything is wrong" meme.

One thing that caught my eye was how the piece highlighted a crowd-sourced data analysis of soccer red cards. The key figure for that analysis is this one:

 

I think the figure and underlying data for this figure are fascinating in that they really highlight the human behavioral variation in data analysis and you can even see some data analysis subcultures emerging from the descriptions of how people did the analysis and justified or not the use of covariates.

One subtlety of the figure that I missed on the original reading is that not all of the estimates being reported are measuring the same thing. For example, if some groups adjusted for the country of origin of the referees and some did not, then the estimates for those two groups are measuring different things (the association conditional on country of origin or not, respectively). In this case the estimates may be different, but entirely consistent with each other, since they are just measuring different things.

If you ask two people to do the analysis and you only ask them the simple question: Are referees more likely to give  red cards to dark skinned players? then you may get a different answer based on those two estimates. But the reality is the answers the analysts are reporting are actually to the questions:

  1. Are referees more likely to give  red cards to dark skinned players holding country of origin fixed?
  2. Are referees more likely to give  red cards to dark skinned players averaging over country of origin (and everything else)?

The subtlety lies in the fact that changes to covariates in the analysis are actually changing the hypothesis you are studying.

So in fact the conclusions in that figure may all be entirely consistent after you condition on asking the same question. I'd be interested to see the same plot, but only for the groups that conditioned on the same set of covariates, for example. This is just one more reason that science is really hard and why I'm so impressed at how well the FiveThirtyEight piece captured this nuance.

 

 

19
Aug

P > 0.05? I can make any p-value statistically significant with adaptive FDR procedures

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

Everyone knows now that you have to correct for multiple testing when you calculate many p-values otherwise this can happen:

http://xkcd.com/882/

 

One of the most popular ways to correct for multiple testing is to estimate or control the false discovery rate. The false discovery rate attempts to quantify the fraction of made discoveries that are false. If we call all p-values less than some threshold t significant, then borrowing notation from this great introduction to false discovery rates 

fdr3

 

So F(t) is the (unknown) total number of null hypotheses called significant and S(t) is the total number of hypotheses called significant. The FDR is the expected ratio of these two quantities, which, under certain assumptions can be approximated by the ratio of the expectations.

 

fdr4

 

To get an estimate of the FDR we just need an estimate for  E[F(t)]  and E[S(t)]. The latter is pretty easy to estimate as just the total number of rejections (the number of p < t). If you assume that the p-values follow the expected distribution then E[F(t)]  can be approximated by multiplying the fraction of null hypotheses, multiplied by the total number of hypotheses and multiplied by t since the p-values are uniform. To do this, we need an estimate for \pi_0, the proportion of null hypotheses. There are a large number of ways to estimate this quantity but it is almost always estimated using the full distribution of computed p-values in an experiment. The most popular estimator compares the fraction of p-values greater than some cutoff to the number you would expect if every single hypothesis were null. This fraction is about the fraction of null hypotheses.

Combining the above equation with our estimates for E[F(t)]  and E[S(t)] we get:

 

fdr5

 

The q-value is a multiple testing analog of the p-value and is defined as:

fdr6

 

This is of course a very loose version of this and you can get a more technical description here. But the main thing to notice is that the q-value depends on the estimated proportion of null hypotheses, which depends on the distribution of the observed p-values. The smaller the estimated fraction of null hypotheses, the smaller the FDR estimate and the smaller the q-value. This suggests a way to make any p-value significant by altering its "testing partners". Here is a quick example. Suppose that we have done a test and have a p-value of 0.8. Not super significant. Suppose we perform this test in conjunction with a number of hypotheses that are null generating a p-value distribution like this.

uniform-pvals

Then you get a q-value greater than 0.99 as you would expect. But if you test that exact same p-value with a ton of other non-null hypotheses that generate tiny p-values in a distribution that looks like this:

significant-pvals

 

Then you get a q-value of 0.0001 for that same p-value of 0.8. The reason is that the estimate of the fraction of null hypotheses goes essentially to zero, which drives down the q-value. You can do this with any p-value, if you make its testing partners have sufficiently low p-values then the q-value will also be as small as you like.

A couple of things to note:

  • Obviously doing this on purpose to change the significance of a calculated p-value is cheating and shouldn't be done.
  • For correctly calculated p-values on a related set of hypotheses this is actually a sensible property to have - if you have almost all very small p-values and one very large p-value, you are doing a set of tests where almost everything appears to be alternative and you should weight that in some sensible way.
  • This is the reason that sometimes a "multiple testing adjusted" p-value (or q-value) is smaller than the p-value itself.
  • This doesn't affect non-adaptive FDR procedures - but those procedures still depend on the "testing partners" of any p-value through the total number of tests performed. This is why people talk about the so-called "multiple testing burden". But that is a subject for a future post. It is also the reason non-adaptive procedures can be severely underpowered compared to adaptive procedures when the p-values are correct.
  • I've appended the code to generate the histograms and calculate the q-values in this post in the following gist.

 

12
Aug

UCLA Statistics 2015 Commencement Address

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

I was asked to speak at the UCLA Department of Statistics Commencement Ceremony this past June. As one of the first graduates of that department back in 2003, I was tremendously honored to be invited to speak to the graduates. When I arrived I was just shocked at how much the department had grown. When I graduated I think there were no more than 10 of us between the PhD and Master's programs. Now they have ~90 graduates per year with undergrad, Master's and PhD. It was just stunning.

Here's the text of what I said, which I think I mostly stuck to in the actual speech.

 

UCLA Statistics Graduation: Some thoughts on a career in statistics

When I asked Rick [Schoenberg] what I should talk about, he said to 'talk for 95 minutes on asymptotic properties of maximum likelihood estimators under nonstandard conditions". I thought this is a great opportunity! I busted out Tom Ferguson’s book and went through my old notes. Here we go. Let X be a complete normed vector space….

I want to thank the department for inviting me here today. It’s always good to be back. I entered the UCLA stat department in 1999, only the second entering class, and graduated from UCLA Stat in 2003. Things were different then. Jan was the chair and there were not many classes so we could basically do whatever we wanted. Things are different now and that’s a good thing. Since 2003, I’ve been at the Department of Biostatistics at the Johns Hopkins Bloomberg School of Public Health, where I was first a postdoctoral fellow and then joined the faculty. It’s been a wonderful place for me to grow up and I’ve learned a lot there.

It’s just an incredible time to be a statistician. You guys timed it just right. I’ve been lucky enough to witness two periods like this, the first time being when I graduated from college at the height of the dot come boom. Today, it’s not computer programming skills that the world needs, but rather it’s statistical skills. I wish I were in your shoes today, just getting ready to startup. But since I’m not, I figured the best thing I could do is share some of the things I’ve learned and talk about the role that these things have played in my own life.

Know your edge: What’s the one thing that you know that no one else seems to know? You’re not a clone—you have original ideas and skills. You might think they’re not valuable but you’re wrong. Be proud of these ideas and use them to your advantage. As an example, I’ll give you my one thing. Right now, I believe the greatest challenge facing the field of statistics today is getting the entire world to know what we in this room already know. Data are everywhere today and the biggest barrier to progress is our collective inability to process and analyze those data to produce useful information. The need for the things that we know has absolutely exploded and we simply have not caught up. That’s why I created, along with Jeff Leek and Brian Caffo, the Johns Hopkins Data Science Specialization, which is currently the most successful massive open online course program ever. Our goal is to teach the entire world statistics, which we think is an essential skill. We’re not quite there yet, but—assuming you guys don’t steal my idea—I’m hopeful that we’ll get there sometime soon.

At some point the edge you have will no longer work: That sounds like a bad thing, but it’s actually good. If what you’re doing really matters, then at some point everyone will be doing it. So you’ll need to find something else. I’ve been confronted with this problem at least 3 times in my life so far. Before college, I was pretty good at the violin, and it opened a lot of doors for me. It got me into Yale. But when I got to Yale, I quickly realized that there were a lot of really good violinists here. Suddenly, my talent didn’t have so much value. This was when I started to pick up computer programming and in 1998 I learned an obscure little language called R. When I got to UCLA I realized I was one of the only people who knew R. So I started a little brown bag lunch series where I’d talk about some feature of R to whomever would show up (which wasn’t many people usually). Picking up on R early on turned out to be really important because it was a small community back then and it was easy to have a big impact. Also, as more and more people wanted to learn R, they’d usually call on me. It’s always nice to feel needed. Over the years, the R community exploded and R’s popularity got to the point where it was being talked about in the New York Times. But now you see the problem. Saying that you know R doesn’t exactly distinguish you anymore, so it’s time to move on again. These days, I’m realizing that the one useful skill that I have is the ability to make movies. Also, my experience being a performer on the violin many years ago is coming in handy. My ability to quickly record and edit movies was one of the key factors that enabled me to create an entire online data science program in 2 months last year.

Find the right people, and stick with them forever. Being a statistician means working with other people. Choose those people wisely and develop a strong relationship. It doesn’t matter how great the project is or how famous or interesting the other person is, if you can’t get along then bad things will happen. Statistics and data analysis is a highly verbal process that requires constant and very clear communication. If you’re uncomfortable with someone in any way, everything will suffer. Data analysis is unique in this way—our success depends critically on other people. I’ve only had a few collaborators in the past 12 years, but I love them like family. When I work with these people, I don’t necessarily know what will happen, but I know it will be good. In the end, I honestly don’t think I’ll remember the details of the work that I did, but I’ll remember the people I worked with and the relationships I built.

So I hope you weren’t expecting a new asymptotic theorem today, because this is pretty much all I’ve got. As you all go on to the next phase of your life, just be confident in your own ideas, be prepared to change and learn new things, and find the right people to do them with. Thank you.

12
Aug

Correlation is not a measure of reproducibility

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

Biologists make wide use of correlation as a measure of reproducibility. Specifically, they quantify reproducibility with the correlation between measurements obtained from replicated experiments. For example, the ENCODE data standards document states

A typical R2 (Pearson) correlation of gene expression (RPKM) between two biological replicates, for RNAs that are detected in both samples using RPKM or read counts, should be between 0.92 to 0.98. Experiments with biological correlations that fall below 0.9 should be either be repeated or explained.

However, for  reasons I will explain here, correlation is not necessarily informative with regards to reproducibility. The mathematical results described below are not inconsequential theoretical details, and understanding them will help you assess new technologies, experimental procedures and computation methods.

Suppose you have collected data from an experiment

x1x2,..., xn

and want to determine if  a second experiment replicates these findings. For simplicity, we represent data from the second experiment as adding unbiased (averages out to 0) and statistically independent measurement error d to the first:

y1=x1+d1, y2=x2+d2, ... yn=xn+dn.

For us to claim reproducibility we want the differences

d1=y1-x1, d2=y2-x2,... ,dn=yn-xn

to be "small". To give this some context, imagine the x and y are log scale (base 2) gene expression measurements which implies the d represent log fold changes. If these differences have a standard deviation of 1, it implies that fold changes of 2 are typical between replicates. If our replication experiment produces measurements that are typically twice as big or twice as small as the original, I am not going to claim the measurements are reproduced. However, as it turns out, such terrible reproducibility can still result in correlations higher than 0.92.

To someone basing their definition of correlation on the current common language usage this may seem surprising, but to someone basing it on math, it is not. To see this, note that the mathematical definition of correlation tells us that because and x are independent:

pearsonformula

This tells us that correlation summarizes the variability of relative to the variability of x. Because of the wide range of gene expression values we observe in practice, the standard deviation of x can easily be as large as 3 (variance is 9). This implies we expect to see correlations as high as 1/sqrt(1+1/9) = 0.95, despite the lack of reproducibility when comparing to y.

Note that using Spearman correlation does not fix this problem. A Spearman correlation of 1 tells us that the ranks of and are preserved, yet doest not summarize the actual differences. The problem comes down to the fact that we care about the variability of and correlation, Pearson or Spearman, does not provide an optimal summary. While correlation relates to the preservation of ranks, a much more appropriate summary of reproducibly is the distance between x and y which is related to the standard deviation of the differences d. A very simple R command you can use to generate this summary statistic is:

sqrt(mean(d^2))

or the robust version:

median(abs(d)) ##multiply by 1.4826 for unbiased estimate of true sd

The equivalent suggestion for plots it to make an MA-plot instead of a scatterplot.

But aren't correlations and distances directly related? Sort of, and this actually brings up another problem. If the x and y are standardized to have average 0 and standard deviation 1 then, yes, correlation and distance are directly related:

distcorr

However, if instead x and y have different average values, which would put into question reproducibility, then distance is sensitive to this problem while correlation is not. If the standard devtiation is 1, the formula is:

 

distcor2

Once we consider units (standard deviations different from 1) then the relationship becomes even more complicated. Two advantages of distance you should be aware of are:

  1. it is in the same units as the data, while correlations have no units making it hard to interpret and select thresholds, and
  2. distance accounts for bias (differences in average), while correlation does not.

A final important point relates to the use of correlation with data that is not approximately normal. The useful interpretation of correlation as a summary statistic stems from the bivariate normal approximation: for every standard unit increase in the first variable, the second variable increased standard units, with the correlation. A  summary of this is here. However, when data is not normal this interpretation no longer holds. Furthermore, heavy tail distributions, which are common in genomics, can lead to instability. Here is an example of uncorrelated data with a single pointed added that leads to correlations close to 1. This is quite common with RNAseq data.

supp_figure_2

 

10
Aug

rafalib package now on CRAN

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

For the last several years I have been collecting functions I routinely use during exploratory data analysis in a private R package. Mike Love and I used some of these in our HarvardX course and now, due to popular demand, I have created man pages and added the rafalib package to CRAN. Mike has made several improvements and added some functions of his own. Here is quick descriptions of the rafalib functions I most use:

mypar - Before making a plot in R I almost always type mypar(). This basically gets around the suboptimal defaults of par. For example, it makes the margins (mar, mpg) smaller and defines RColorBrewer colors as defaults.  It is optimized for the RStudio window. Another advantage is that you can type mypar(3,2) instead of par(mfrow=c(3,2)). bigpar() is optimized for R presentations or PowerPoint slides.

as.fumeric - This function turns characters into factors and then into numerics. This is useful, for example, if you want to plot values x,y with colors defined by their corresponding categories saved in a character vector labsplot(x,y,col=as.fumeric(labs)).

shist (smooth histogram, pronounced shitz) - I wrote this function because I have a hard time interpreting the y-axis of density. The height of the curve drawn by shist can be interpreted as the height of a histogram if you used the units shown on the plot. Also, it automatically draws a smooth histogram for each entry in a matrix on the same plot.

splot (subset plot) - The datasets I work with are typically large enough that
plot(x,y) involves millions of points, which is a problem. Several solution are available to avoid over plotting, such as alpha-blending, hexbinning and 2d kernel smoothing. For reasons I won't explain here, I generally prefer subsampling over these solutions. splot automatically subsamples. You can also specify an index that defines the subset.

sboxplot (smart boxplot) - This function draws points, boxplots or outlier-less boxplots depending on sample size. Coming soon is the kaboxplot (Karl Broman box-plots) for when you have too many boxplots.

install_bioc - For Bioconductor users, this function simply does the source("http://www.bioconductor.org/biocLite.R") for you and then uses BiocLite to install.

unnamed

09
Aug

Interested in analyzing images of brains? Get started with open access data.

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone
Editor's note: This is a guest post by Ani Eloyan. She is an Assistant Professor of Biostatistics at Brown University. Dr. Eloyan’s work focuses on semi-parametric likelihood based methods for matrix decompositions, statistical analyses of brain images, and the integration of various types of complex data structures for analyzing health care data. She received her PhD in statistics from North Carolina State University and subsequently completed a postdoctoral fellowship in the Department of Biostatistics at Johns Hopkins University. Dr. Eloyan and her team won the ADHD200 Competition discussed in this article. She tweets @eloyan_ani.
 
Neuroscience is one of the exciting new fields for biostatisticians interested in real world applications where they can contribute novel statistical approaches. Most research in brain imaging has historically included studies run for small numbers of patients. While justified by the costs of data collection, the claims based on analyzing data for such small numbers of subjects often do not hold for our populations of interest. As discussed in this article, there is a huge demand for biostatisticians in the field of quantitative neuroscience; so called neuroquants or neurostatisticians. However, while more statisticians are interested in the field, we are far from competing with other substantive domains. For instance, a quick search of abstract keywords in the online program of the upcoming JSM2015 conference of “brain imaging” and “neuroscience” results in 15 records, while a search of the words “genomics” and “genetics” generates 76 records.
Assuming you are trained in statistics and an aspiring neuroquant, how would you go about working with brain imaging data? As a graduate student in the Department of Statistics at NCSU several years ago, I was very interested in working on statistical methods that would be directly applicable to solve problems in neuroscience. But I had this same question: “Where do I find the data?” I soon learned that to reallyapproach substantial relevant problems I also needed to learn about the subject matter underlying these complex data structures.
In recent years, several leading groups have uploaded their lab data with the common goal of fostering the collection of high dimensional brain imaging data to build powerful models that can give generalizable results. Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC) founded in 2006 is a platform for public data sharing that facilitates streamlining data processing pipelines and compiling high dimensional imaging datasets for crowdsourcing the analyses. It includes data for people with neurological diseases and neurotypical children and adults. If you are interested in Alzheimer’s disease, you can check out ADNI. ABIDE provides data for people with Autism Spectrum Disorder and neurotypical peers. ADHD200 was released in 2011 as a part of a competition to motivate building predictive methods for disease diagnoses using functional magnetic resonance imaging (MRI) in addition to demographic information to predict whether a child has attention deficit hyperactivity disorder (ADHD). While the competition ended in 2011, the dataset has been widely utilized afterwards in studies of ADHD.  According to Google Scholar, the paper introducing the ABIDE set has been cited 129 times since 2013 while the paper discussing the ADHD200 has been cited 51 times since 2012. These are only a few examples from the list of open access datasets that could of utilized by statisticians. 
Anyone can download these datasets (you may need to register and complete some paperwork in some cases), however, there are several data processing and cleaning steps to perform before the final statistical analyses. These preprocessing steps can be daunting for a statistician new to the field, especially as the tools used for preprocessing may not be available in R. This discussion makes the case as to why statisticians need to be involved in every step of preprocessing the data, while this R package contains new tools linking R to a commonly used platform FSL. However, as a newcomer, it can be easier to start with data that are already processed. This excellent overview by Dr. Martin Lindquist provides an introduction to the different types of analyses for brain imaging data from a statisticians point of view, while ourpaper provides tools in R and example datasets for implementing some of these methods. At least one course on Coursera can help you get started with functional MRI data. Talking to and reading the papers of biostatisticians working in the field of quantitative neuroscience and scientists in the field of neuroscience is the key.
09
Aug

Statistical Theory is our "Write Once, Run Anywhere"

Tweet about this on TwitterShare on FacebookShare on Google+Share on LinkedInEmail this to someone

Having followed the software industry as a casual bystander, I periodically see the tension flare up between the idea of writing "native apps", software that is tuned to a particular platform (Windows, Mac, etc.) and more cross-platform apps, which run on many platforms without too much modification. Over the years it has come up in many different forms, but they fundamentals are the same. Back in the day, there was Java, which was supposed to be the platform that ran on any computing device. Sun Microsystems originated the phrase "Write Once, Run Anywhere" to illustrate the cross-platform strengths of Java. More recently, Steve Jobs famously banned Flash from any iOS device. Apple is also moving away from standards like OpenGL and towards its own Metal platform.

What's the problem with "write once, run anywhere", or of cross-platform development more generally, assuming it's possible? Well, there are a number of issues: often there are performance penalties, it may be difficult to use the native look and feel of a platform, and you may be reduced to using the "lowest common denominator" of feature sets. It seems to me that anytime a new meta-platform comes out that promises to relieve programmers of the burden of having to write for multiple platforms, it eventually gets modified or subsumed by the need to optimize apps for a given platform as much as possible. The need to squeeze as much juice out of an app seems to be too important an opportunity to pass up.

In statistics, theory and theorems are our version of "write once, run anywhere". The basic idea is that theorems provide an abstract layer (a "virtual machine") that allows us to reason across a large number of specific problems. Think of the central limit theorem, probably our most popular theorem. It could be applied to any problem/situation where you have a notion of sample size that could in principle be increasing.

But can it be applied to every situation, or even any situation? This might be more of a philosophical question, given that the CLT is stated asymptotically (maybe we'll find out the answer eventually). In practice, my experience is that many people attempt to apply it to problems where it likely is not appropriate. Think, large-scale studies with a sample size of 10. Many people will use Normal-based confidence intervals in those situations, but they probably have very poor coverage.

Because the CLT doesn't apply in many situations (small sample, dependent data, etc.), variations of the CLT have been developed, as well as entirely different approaches to achieving the same ends, like confidence intervals, p-values, and standard errors (think bootstrap, jackknife, permutation tests). While the CLT an provide beautiful insight in a large variety of situations, in reality, one must often resort to a custom solution when analyzing a given dataset or problem. This should be a familiar conclusion to anyone who analyzes data. The promise of "write once, run anywhere" is always tantalizing, but the reality never seems to meet that expectation.

Ironically, if you look across history and all programming languages, probably the most "cross-platform" language is C, which was originally considered to be too low-level to be broadly useful. C programs run on basically every existing platform and the language has been completely standardized so that compilers can be written to produce well-defined output. The keys to C's success I think are that it's a very simple/small language which gives enormous (sometimes dangerous) power to the programmer, and that an enormous toolbox (compiler toolchains, IDEs) has been developed over time to help developers write applications on all platforms.

In a sense, we need "compilers" that can help us translate statistical theory for specific data analysis problems. In many cases, I'd imagine the compiler would "fail", meaning the theory was not applicable to that problem. This would be a Good Thing, because right now we have no way of really enforcing the appropriateness of a theorem for specific problems.

More practically (perhaps), we could develop data analysis pipelines that could be applied to broad classes of data analysis problems. Then a "compiler" could be employed to translate the pipeline so that it worked for a given dataset/problem/toolchain.

The key point is to recognize that there is a "translation" process that occurs when we use theory to justify certain data analysis actions, but this translation process is often not well documented or even thought through. Having an explicit "compiler" for this would help us to understand the applicability of certain theorems and may serve to prevent bad data analysis from occurring.