Negative controls

Loading...

Del curso dictado por Johns Hopkins University

Data Science in Real Life

1322 calificaciones

Johns Hopkins University

Data Science in Real Life

1322 calificaciones

Curso 4 de 5 en SpecializationExecutive Data Science

Have you ever had the perfect data science experience? The data pull went perfectly. There were no merging errors or missing data. Hypotheses were clearly defined prior to analyses. Randomization was performed for the treatment of interest. The analytic plan was outlined prior to analysis and followed exactly. The conclusions were clear and actionable decisions were obvious. Has that every happened to you? Of course not. Data analysis in real life is messy. How does one manage a team facing real data analyses? In this one-week course, we contrast the ideal with what happens in real life. By contrasting the ideal, you will learn key concepts that will help you manage real life analyses. This is a focused course designed to rapidly get you up to speed on doing data science in real life. Our goal was to make this as convenient as possible for you without sacrificing any essential content. We've left the technical information aside so that you can focus on managing your team and moving it forward. After completing this course you will know how to: 1, Describe the “perfect” data science experience 2. Identify strengths and weaknesses in experimental designs 3. Describe possible pitfalls when pulling / assembling data and learn solutions for managing data pulls. 4. Challenge statistical modeling assumptions and drive feedback to data analysts 5. Describe common pitfalls in communicating data analyses 6. Get a glimpse into a day in the life of a data analysis manager. The course will be taught at a conceptual level for active managers of data scientists and statisticians. Some key concepts being discussed include: 1. Experimental design, randomization, A/B testing 2. Causal inference, counterfactuals, 3. Strategies for managing data quality. 4. Bias and confounding 5. Contrasting machine learning versus classical statistical inference Course promo: https://www.youtube.com/watch?v=9BIYmw5wnBI Course cover image by Jonathan Gross. Creative Commons BY-ND https://flic.kr/p/q1vudb

De la lección

Introduction, the perfect data science experience

This course is one module, intended to be taken in one week. Please do the course roughly in the order presented. Each lecture has reading and videos. Except for the introductory lecture, every lecture has a 5 question quiz; get 4 out of 5 or better on the quiz.

Multiplicity6:52

Effect size, significance, & modeling7:31

Comparison with benchmark effects4:09

Negative controls5:27

Conoce a los instructores

Brian Caffo, PhD
Professor, Biostatistics
Bloomberg School of Public Health
Jeff Leek, PhD
Associate Professor, Biostatistics
Bloomberg School of Public Health
Roger D. Peng, PhD
Associate Professor, Biostatistics
Bloomberg School of Public Health

Hi.

This is the lecture on negative controls.

So this is part of our discussion of looking at statistically significant or

non-significant results and adding meat to our hypothesis testing.

And in this case, what we're concerned with is the problem of,

did the process that we executed the hypothesis in,

was that what created the significance, or was it a real intrinsically true effect?

Okay.

So let's restate the problem.

You're worried that your results are more due to process than a real effect.

Well how do you check?

Well there's a lot of ways and a lot of the mechanisms of statistical significance

are trying to check against that.

They're evaluating your results relative to the uncertainty in your data and so on.

But this, what I'm about to present is a very practical solution to that.

Just a real sort of more data science-y way to think about it.

And the idea is to perform a negative control.

And you're basically going to repeat the study for

a variable that is known to have no association.

So, let me give you a very famous example in the area that I work in.

What everyone is talking about in

brain activation studies now, is in this idea of connectivity.

That two different areas of the brain,

their brain activity tends to track with one another.

And that's the idea of so-called connectivity.

It's interesting to note that what someone did at one point

is they used all these tools that people like me and

other people that work in the area have developed, and they looked at them

in an area of the brain where there is no brain, the ventricles, okay?

So the in the ventricles there's nothing but cerebral spinal fluid, so

there's nothing there.

And for two locations in the ventricles,

they used the same techniques that we use for two locations in the brain

to study correlations or connectivity as we like to call it.

And what they found is significant results.

And that's an interesting finding cuz there's no brain there, so

it's not brain conductivity they're finding, it purely has to do with process.

And what they found out is that this was due to a couple of interesting

artifactual components of the data.

One, it was due to sort of head motion.

As people move their head, it created similar patterns in the two locations in

this area of the ventricles and that caused correlations.

Those effects permeated the rest of the image, so a lot of

what people were reporting as connectivity was really due to head motion.

It's also true that as we breathe, and as our heart beats,

that changes the character of the signal dynamically and

those were also related to creating spurious correlations.

The broader point, though, is that what these investigators did

is they repeated an analysis that everyone was excited about.

Everyone was excited about finding remote correlations in

the brain that were putatively brain networks and

understanding how they were related to disease and stuff like that.

However, what these researchers did is,

they said okay let's just study the process by looking at an instance

where we know that there can't be an effect, but in every other sense,

we've replicated the analysis in exactly the same way and they found an effect.

So these kinds of negative controls offer a very simple way to investigate spurious

effects due to things like confounding or multiplicity.

In this case, it was confounding.

The correlation between this two voxels is actually confounded by head motion, for

example.

So what are the characteristics of a good negative control?

Well, they're variables that are realistic but known to have no association.

So in this case, they had a perfect one.

They had something that was in the image already, that they could look at,

it was subject to all the same processing experience that the rest of

the brain got and so this was a good version of a negative control.

If there's one criticism of negative control, it's often very hard to

find something where you know for sure there can't possibly be an effect.

So another strategy that people employ rather than

executing things like negative controls, is they do things like permutation tests.

They actually formally break the association by permuting one of

the variables so that there can't be an association, they've broken it.

So that's a little bit more of an advance of a topic, but the idea is quite similar.

So this is a simple technique.

And if you're feeling queasy about

a process that one of the people you're managing went through, a very long,

complicated set of data munging and analysis and several models being built,

ask them to just repeat the whole process for something that is,

on face value, can't possibly exhibit an association, and see what comes out.

And it's a great way to offer a sanity check,

and to check influences due to process.

Explora nuestro catálogo

Inscríbete de manera gratuita y obtén recomendaciones personalizadas, actualizaciones y ofertas.

Coursera brinda acceso universal a la mejor educación del mundo, al asociarse con las mejores universidades y organizaciones, para ofrecer cursos en línea.

© 2018 Coursera Inc. Todos los derechos reservados.

Descargar en la App Store

Consíguelo en Google Play