Master's in Data Science before Clinical Psych PhD?

[ethospathoslegos]

Hi everyone,
I'm a few years out from undergrad and currently working as a research coordinator in a psychiatry department, looking to ultimately complete a PhD in clinical psychology (ideally with a focus on neuropsych/pharmacology). My undergrad GPA is about a 3.0 (pre-med classes "for fun" and undiagnosed ADHD, mistakes were made). From lurking in this forum and talking to mentors in the field, it seems like I will need to do a master's degree before applying to PhD programs in order to be competitive.

Depending on an applicant's weaknesses, most of the threads on this forum seem to recommend clinical master's degrees (like an MSW) or research-oriented psychology master's degrees designed for those looking to become more competitive for graduate programs (like this MA in Psychology in Education from Columbia). I'd like to complete a master's that has value as a terminal degree, and I think I have enough research experience to be competitive (5 years full-time after college, 4 years during college, a handful of publications and posters). I currently do most of the data management for my group, and I've audited some graduate courses in data science that I enjoyed. If I were to complete a master's in data science or statistics & data science, would this strengthen my application for PhD programs?

My thinking is that these master's programs provide training in statistics, managing large data sets, and programming (I do MRI research so this is useful), all of which would be valuable in a PhD program. I haven't seen anyone discuss data science as a possibility though, and I'm wondering if there's a reason for that that I'm overlooking. Thanks!

---

Members don't see this ad.

 

[R. Matey]

I have a hard time seeing the value of a master's degree in data science only because the skills can be learned so easily and much more cheaply through any number of online bootcamps or tutorials, some of which are completely free. For instance, the jamovi project has a free stats program that helps teach people to learn to code in R. I've done some work with a group of data scientists and many of them have fairly diverse backgrounds academically (e.g.: epidemiology, environmental science, geography) and acquired their skills during their degree programs or through such means. And much of what you learn in a master's degree program in statistics you also learn in a Ph.D. program if you pursue it albeit there might be more formal coursework in the master's program.

My own circuitous route to becoming a psychologist involved a clinical master's degree, which I don't recommend right now. These paths, for better or worse, are closing off as the helping professions become more siloed from one another. The master's program at Columbia looks interesting and could be a good fit depending on what your ultimate goals are and whether you felt you could work with the neuropsych mentor.

 

[WisNeuro]

I have a hard time seeing the value of a master's degree in data science only because the skills can be learned so easily and much more cheaply through any number of online bootcamps or tutorials, some of which are completely free. For instance, the jamovi project has a free stats program that helps teach people to learn to code in R. I've done some work with a group of data scientists and many of them have fairly diverse backgrounds academically (e.g.: epidemiology, environmental science, geography) and acquired their skills during their degree programs or through such means. And much of what you learn in a master's degree program in statistics you also learn in a Ph.D. program if you pursue it albeit there might be more formal coursework in the master's program.

My own circuitous route to becoming a psychologist involved a clinical master's degree, which I don't recommend right now. These paths, for better or worse, are closing off as the helping professions become more siloed from one another. The master's program at Columbia looks interesting and could be a good fit depending on what your ultimate goals are and whether you felt you could work with the neuropsych mentor.

As a stepping stone to a doctoral program, yeah, doesn't really help much. But, as a standalone fallback degree, it has value. Plenty of data science jobs which require the degree.

 

[summerbabe]

My thinking is that these master's programs provide training in statistics, managing large data sets, and programming (I do MRI research so this is useful), all of which would be valuable in a PhD program.

All of this can be useful for a PhD. However, will this degree help further demonstrate your fit for this field and programs that you would apply to?

My dissertation ended up using structural equation modeling run in Amos for its analysis. When I started my PhD, I had zero familiarity with either but I did have concrete directions for research based on previous projects/exposure so as my dissertation idea developed, I ended up taking a class on SEM and then worked with my mentor and a quant expert at my school on the analytical piece.

I think many of us learn technical skills along the way and since you’re probably not quantitatively challenged, I would focus my attention toward gaining more psychology-specific experiences that a clinical psych PI/program would see as good foundational experiences which suggest you’d likely be successful in grad school.

 

[PsychPhDone]

Hi everyone,
I'm a few years out from undergrad and currently working as a research coordinator in a psychiatry department, looking to ultimately complete a PhD in clinical psychology (ideally with a focus on neuropsych/pharmacology). My undergrad GPA is about a 3.0 (pre-med classes "for fun" and undiagnosed ADHD, mistakes were made). From lurking in this forum and talking to mentors in the field, it seems like I will need to do a master's degree before applying to PhD programs in order to be competitive.

Depending on an applicant's weaknesses, most of the threads on this forum seem to recommend clinical master's degrees (like an MSW) or research-oriented psychology master's degrees designed for those looking to become more competitive for graduate programs (like this MA in Psychology in Education from Columbia). I'd like to complete a master's that has value as a terminal degree, and I think I have enough research experience to be competitive (5 years full-time after college, 4 years during college, a handful of publications and posters). I currently do most of the data management for my group, and I've audited some graduate courses in data science that I enjoyed. If I were to complete a master's in data science or statistics & data science, would this strengthen my application for PhD programs?

My thinking is that these master's programs provide training in statistics, managing large data sets, and programming (I do MRI research so this is useful), all of which would be valuable in a PhD program. I haven't seen anyone discuss data science as a possibility though, and I'm wondering if there's a reason for that that I'm overlooking. Thanks!

What's your undergrad psych gpa?

 

[R. Matey]

As a stepping stone to a doctoral program, yeah, doesn't really help much. But, as a standalone fallback degree, it has value. Plenty of data science jobs which require the degree.

Right, but that's also like saying an undergrad computer science degree is needed to work in Big Tech. I have a few friends working at name brand companies and they'll tell you the educational backgrounds of folks working there doing the same or similar types of work vary widely. Some people have graduate degrees in the field, others never went to college. It's anecdotal, but I would expect as much in industry where they care much more about the skillset than the degree title.

The other issue is "data scientist" is a very loosely defined term partly because it's still emerging. Almost everyone and their cat can claim it if they have coding experience.

 

[borne_before]

I'm going to counter some of what others' have said.

First, as a homie with ADHD, I also had to get masters before becoming competitive for a doctorate. I got my masters in educational psychology because I found test development and research methodology so interesting. But that masters was largely useless and realized early that I wanted to work with people instead of test development. So I applied to school psych programs and got in.

Second, if you love data, research, and other quant stuff, then absolutely pursue it. The times are a changin' and the era of big data is here. Many university based clinical psych programs will not consider you if you do not have interest in research. In fact, a lot of my buddies in the clinical psych programs had no intention of conducting research after grad school - but they still had to give it much lip service. This emphasis is seen on alternative accreditation movements like PCSAS and the latest paper linked here about making you able to a PhD in clinical psychology without completing an internship. The masters in will absolutely give you value as part of a research lab as the data/coding slave. The majority of my citations come from being the data slave and doing the stats.

Third, bully if you can get into a masters and have your tuition subsidized as part of your research group (basically get a masters for as cheap as possible, and being a university employee will help that).

Fourth, your masters degree will change you. For instance, you may look for another degree in cog science or educational economics or whatever. The goal of getting a phd in psychology may change. Let it. I feel like my experience in my masters was some of my best conceptual work. It was fun!

Fif, A masters degree with a good GPA, good letters of rec, strong relationships developed, etc., will show perspective programs that you can do it!

 

[ethospathoslegos]

What's your undergrad psych gpa?

My undergrad psych GPA is 3.76.

 

[Ollie123]

Can't speak for other labs, but I would absolutely LOVE to get someone with a master's in data science into my lab and such a person would be given extreme preference over a generic psychology master's degree as long as the rest of their background was adequate and they had a compelling reason for doing so. The degree alone isn't going to be enough, but if you get the degree while working in a behavior-oriented lab, do a mental-health related thesis, etc. I would consider this a slam dunk for positions in my lab. Heck, I've toyed with the idea of getting a data science master's myself. I have a bajillion-and-one projects I'd love to run involving data science techniques no psychology program offers courses in and most psychologists may not have even heard of so my only rate-limiting factor here is time and analytic expertise (as I lack the time to develop the expertise myself!).

I do pretty technically-oriented research (incl. MRI) compared to many though. I think for the right lab it would be a spectacular fit, but for a prototypical lab a bit less so. If the end goal is definitely to be a practicing clinician I wouldn't go this route, but if you are committed to a given research topic I think its a great idea.

 

[ethospathoslegos]

All of this can be useful for a PhD. However, will this degree help further demonstrate your fit for this field and programs that you would apply to?

My dissertation ended up using structural equation modeling run in Amos for its analysis. When I started my PhD, I had zero familiarity with either but I did have concrete directions for research based on previous projects/exposure so as my dissertation idea developed, I ended up taking a class on SEM and then worked with my mentor and a quant expert at my school on the analytical piece.

I think many of us learn technical skills along the way and since you’re probably not quantitatively challenged, I would focus my attention toward gaining more psychology-specific experiences that a clinical psych PI/program would see as good foundational experiences which suggest you’d likely be successful in grad school.

Thank you for your response! I've received feedback from a few posters that, while a data science/statistics degree might be helpful as a terminal degree for a data science career, it won't necessarily make me competitive for a clinical psych graduate program. I'm curious, do you feel that there is a better masters program to complete that would also have value as a terminal degree?

I'm open to completing a clinical masters degree (like an MSW), but another poster in this thread discouraged this. My biggest hesitation with many of the masters programs that are designed to prepare students for PhD programs is that they are costly, unfunded degrees that don't seem to have professional value except as a stepping stone for a PhD program (though perhaps I malign them unfairly). If it's not realistic to expect to complete a masters that makes one competitive for a PhD program and has standalone value, though, I appreciate receiving that feedback now. Thank you again for taking the time to respond to my question.

 

[Dazen]

Can't speak for other labs, but I would absolutely LOVE to get someone with a master's in data science into my lab and such a person would be given extreme preference over a generic psychology master's degree as long as the rest of their background was adequate and they had a compelling reason for doing so. The degree alone isn't going to be enough, but if you get the degree while working in a behavior-oriented lab, do a mental-health related thesis, etc. I would consider this a slam dunk for positions in my lab. Heck, I've toyed with the idea of getting a data science master's myself. I have a bajillion-and-one projects I'd love to run involving data science techniques no psychology program offers courses in and most psychologists may not have even heard of so my only rate-limiting factor here is time and analytic expertise (as I lack the time to develop the expertise myself!).

I do pretty technically-oriented research (incl. MRI) compared to many though. I think for the right lab it would be a spectacular fit, but for a prototypical lab a bit less so. If the end goal is definitely to be a practicing clinician I wouldn't go this route, but if you are committed to a given research topic I think its a great idea.

What data science techniques are you most interested in or see the most value associated with? I'm very excited about math and trying to figure out what to study in my "spare" time (currently learning SEM)

 

[summerbabe]

I'm curious, do you feel that there is a better masters program to complete that would also have value as a terminal degree?

I’m gonna give you the ‘it depends’ answer, in part because PhD programs differ and mentorship-based programs (e.g., where a PI has near sole discretion to make offers) will look for different things so what is seen as a good stepping stone will differ.

But it’s pretty standard that successful PhD applicants have been a part of a research team/lab and contributed to original, peer-reviewed research in some way. It’s also possible to get that experience as a postbac without enrolling in a program through unpaid volunteering or as a paid RA.

If you are looking for a program to be a stepping stone, you should probably avoid things like MSW or MFT degrees since they aren’t in the psychology field and generally don’t emphasize or require original research.

Some MS programs with a research focus (including ones in departments that also house PhD programs and share coursework) will also result in being eligible to apply for a LPC or LMHC license, in case you end up deciding against or aren’t able to pursue a PhD.

Again, this is very general and your specific situation should dictate the best path for you, which could include data science training.

 

[ethospathoslegos]

I have a hard time seeing the value of a master's degree in data science only because the skills can be learned so easily and much more cheaply through any number of online bootcamps or tutorials, some of which are completely free. For instance, the jamovi project has a free stats program that helps teach people to learn to code in R. I've done some work with a group of data scientists and many of them have fairly diverse backgrounds academically (e.g.: epidemiology, environmental science, geography) and acquired their skills during their degree programs or through such means. And much of what you learn in a master's degree program in statistics you also learn in a Ph.D. program if you pursue it albeit there might be more formal coursework in the master's program.

My own circuitous route to becoming a psychologist involved a clinical master's degree, which I don't recommend right now. These paths, for better or worse, are closing off as the helping professions become more siloed from one another. The master's program at Columbia looks interesting and could be a good fit depending on what your ultimate goals are and whether you felt you could work with the neuropsych mentor.

Hey there, thank you for your feedback. You are certainly correct that data science skills can be learned outside of a graduate program (I developed my interest in data science by completing online tutorials and MOOCs). My thought process regarding pursuing a master's in data science was that it would demonstrate my ability to (hopefully) receive good grades while pursuing challenging graduate coursework that is somewhat relevant to a PhD program. However, it seems like this wouldn't really be relevant enough to enhance my application to a clinical psych PhD program. If you don't mind my asking a few follow-up questions, could you expand a bit more on why you don't recommend a clinical master's degree? Are there any master's degrees that you would recommend? My ultimate goal would be to work in research.

 

[R. Matey]

Hey there, thank you for your feedback. You are certainly correct that data science skills can be learned outside of a graduate program (I developed my interest in data science by completing online tutorials and MOOCs). My thought process regarding pursuing a master's in data science was that it would demonstrate my ability to (hopefully) receive good grades while pursuing challenging graduate coursework that is somewhat relevant to a PhD program. However, it seems like this wouldn't really be relevant enough to enhance my application to a clinical psych PhD program. If you don't mind my asking a few follow-up questions, could you expand a bit more on why you don't recommend a clinical master's degree? Are there any master's degrees that you would recommend? My ultimate goal would be to work in research.

Yeah, especially because they are skills you can learn while in you're in graduate school. It would be better to demonstrate a research interest with a compatible mentor with where you are. To @Ollie123's point, machine learning techniques are currently on the cutting edge in psychological science so they are desirable, but it seems you already have a bit of competence going in and I personally being a content expert is what separates you from being just another numbers person. Speaking for myself, I have a datacamp subscription, which has been very useful to me. Andrew Ng's course on Machine Learning is also free on Coursera. These are both far cheaper than a master's degree

Of course. The issue with clinical master's degree en route to Ph.D. programs is the programs usually do not do a very decent job prepping you to do research. It's no slight to them, they only have two or three years to teach you some cursory understanding of clinical psychology and to be, at the very least, not a harmful clinician. That usually comes at the expense of preparing you for a research career. If that is what you want, you're better off in a research based master's program like the one you listed or a general psychology master's program that is research intensive. You'll get a good foundation for the skills you would like anyways and be able to show a trajectory towards a research mentor's program.

 

[Ollie123]

What data science techniques are you most interested in or see the most value associated with? I'm very excited about math and trying to figure out what to study in my "spare" time (currently learning SEM)

A little like asking whether a screwdriver is more valuable than a hammer. They're tools, each is suited for a different purpose and what is better for a given project I might pursue could be utterly useless for other purposes. For what its worth, I would consider SEM to fall under the umbrella of traditional biostats/psych stats given at its core it is basically just the marriage of factor analysis and regression. I'd actually consider it intermediate at best at this point as stats goes and virtually any science-focused psychology program should have tons of people with expertise in it. So it isn't really what I mean when I say data science.

Random examples:
Natural language processing for EHR extraction/analysis
Computer vision techniques (image classification via deep CNN/SVO, object detection, semantic segmentation) to study environments
Pattern recognition for imaging and EEG data data (MVPA, MPCA, dynamic time warping, etc.)
Hidden markov techniques and kriging for studying activity patterns

Like I said they are tools and ones we aren't using very often in our field right now. Name a random ML technique and I can probably give you a half dozen ways it could be used.

 

[Dazen]

A little like asking whether a screwdriver is more valuable than a hammer. They're tools, each is suited for a different purpose and what is better for a given project I might pursue could be utterly useless for other purposes. For what its worth, I would consider SEM to fall under the umbrella of traditional biostats/psych stats given at its core it is basically just the marriage of factor analysis and regression. I'd actually consider it intermediate at best at this point as stats goes and virtually any science-focused psychology program should have tons of people with expertise in it. So it isn't really what I mean when I say data science.

Random examples:
Natural language processing for EHR extraction/analysis
Computer vision techniques (image classification via deep CNN/SVO, object detection, semantic segmentation) to study environments
Pattern recognition for imaging and EEG data data (MVPA, MPCA, dynamic time warping, etc.)
Hidden markov techniques and kriging for studying activity patterns

Like I said they are tools and ones we aren't using very often in our field right now. Name a random ML technique and I can probably give you a half dozen ways it could be used.

Thanks so much for all the info! Sorry that my question wasn't more specific. I just started my first year in a Clinical PhD program and have substantial math/stats experience and interest. I've asked around within the department about techniques to learn (most answers have been SEM, conditional process analysis, and potentially network analysis, so basically traditional techniques), because I am aware of how much I don't know. I feel like I'm still at the step where I don't know enough about all the possibilities to decide which would be most relevant for potential research interests, especially in the data science world, but I really want to start getting that basic foundation and some more specific knowledge/skills. So I guess my question is how do I figure out where to start?

Happy to put this in a new thread if it's too off-topic or chat in PMs. I appreciate your perspective and strong quantitative and research focus!

 

[Ollie123]

Thanks so much for all the info! Sorry that my question wasn't more specific. I just started my first year in a Clinical PhD program and have substantial math/stats experience and interest. I've asked around within the department about techniques to learn (most answers have been SEM, conditional process analysis, and potentially network analysis, so basically traditional techniques), because I am aware of how much I don't know. I feel like I'm still at the step where I don't know enough about all the possibilities to decide which would be most relevant for potential research interests, especially in the data science world, but I really want to start getting that basic foundation and some more specific knowledge/skills. So I guess my question is how do I figure out where to start?

Happy to put this in a new thread if it's too off-topic or chat in PMs. I appreciate your perspective and strong quantitative and research focus!

Feel free to start another thread but fair warning that this isn't really a question anyone can answer for you. Research topic should dictate the techniques, not vice versa. Again - tool analogy. No one can tell you whether it is better to learn to use a saw or a hammer without knowing what you are trying to do. I might recommend something that is great for multivoxel pattern analysis, but if you have no experience in MRI, have no interest in MRI or plans to do MRI it would be completely and utterly useless.

I'd focus on narrowing your content area and learning very generally the scope of what is out there. At this stage, you don't need to know how to do XYZ but even knowing what the various techniques involve will open the door to lots of new questions since they help you frame new questions.

 

[Dazen]

Feel free to start another thread but fair warning that this isn't really a question anyone can answer for you. Research topic should dictate the techniques, not vice versa. Again - tool analogy. No one can tell you whether it is better to learn to use a saw or a hammer without knowing what you are trying to do. I might recommend something that is great for multivoxel pattern analysis, but if you have no experience in MRI, have no interest in MRI or plans to do MRI it would be completely and utterly useless.

I'd focus on narrowing your content area and learning very generally the scope of what is out there. At this stage, you don't need to know how to do XYZ but even knowing what the various techniques involve will open the door to lots of new questions since they help you frame new questions.

Thanks so much for the clarity. I have a decently narrowed content area (at least I have an approved master’s thesis topic), so I’d like to focus more on learning what techniques are out there, so that I can ask better research questions in the future. I totally agree with the idea that, without even knowing what techniques are out there, it’s difficult to know what types of questions I could even answer. Sorry to ask another question, but do you know any good resources that give an overview of a bunch of different data science techniques, so that I can have a general idea of the landscape? General enthusiasm is still there, but I want to make sure I point it in the right direction.

 

[Ollie123]

Honestly, no. Believe me, I've been looking for one too. I kinda get why though, given this is an incredibly broad topic with disparate fields and wildly disparate applications. If we tried to create an ultimate source of "methods" used in psychology research, it would be a catastrophic failure. Same thing.

Best advice I have is to take some intro courses on coursera, venture down the rabbit hole that is wikipedia and read voraciously on the topic, including topics that have nothing to do with psychology (general medicine, science, nature, etc.). It was a dermatology paper that first piqued our interest in this and spurred us to seek out new collaborators (Dermatologist-level classification of skin cancer with deep neural networks - Nature). Watch the movie Alpha Go if you haven't yet and realize that its core, the math essentially distills down to giving a computer a dopamine system.

Lastly, recognize this is going to be tough to REALLY dig in on graduate school and you probably won't get to do anything truly crazy/innovative for a thesis/dissertation and that's OK. Tradition in psychology - for understandable reasons - is you collect your data and run your own analyses. These things require an obscene amount of data that is often untenable for a student project. I don't know your background, but chances of you actually being able to do much beyond the simplest analysis yourself is relatively slim. Any schmuck psychology grad student can run a factor analysis or even a fairly complex random effects model. Good ones can implement fancy latent growth models or computational cognition models. Could you figure out how to run a lasso regression? Sure. Could you implement a multilayer perceptron classifier embedded inside a deep neural network that scaffolds off Inception while folding in a Gaussian Process model? I've been coding since I was 10, darn near went into computer science instead of psychology, was the stats geek in grad school and I might retire before I got that done successfully. This kind of stuff we have had multiple PhD-level engineers working for months to implement analysis for a single research question. So even if you "could" do it yourself....you won't because it isn't an efficient use of time. I think it is very much a bad thing that virtually all senior psychologists move away from being hands-on with their own data as analysis gets offloaded to students, post-docs and statisticians. I'm fighting tooth and nail to avoid becoming one and its still happening. In this context though, it really is necessary. I say all this just to make the point that for the really high-end stuff you really shouldn't focus too much on learning to implement the analysis. Focus on learning what it can do, where to find people who know how to implement it and how to talk to them and strategize about it because this is definitely not like changing the variable names in some SPSS syntax.

 

[R. Matey]

Lastly, recognize this is going to be tough to REALLY dig in on graduate school and you probably won't get to do anything truly crazy/innovative for a thesis/dissertation and that's OK. Tradition in psychology - for understandable reasons - is you collect your data and run your own analyses. These things require an obscene amount of data that is often untenable for a student project. I don't know your background, but chances of you actually being able to do much beyond the simplest analysis yourself is relatively slim. Any schmuck psychology grad student can run a factor analysis or even a fairly complex random effects model. Good ones can implement fancy latent growth models or computational cognition models. Could you figure out how to run a lasso regression? Sure. Could you implement a multilayer perceptron classifier embedded inside a deep neural network that scaffolds off Inception while folding in a Gaussian Process model? I've been coding since I was 10, darn near went into computer science instead of psychology, was the stats geek in grad school and I might retire before I got that done successfully. This kind of stuff we have had multiple PhD-level engineers working for months to implement analysis for a single research question. So even if you "could" do it yourself....you won't because it isn't an efficient use of time. I think it is very much a bad thing that virtually all senior psychologists move away from being hands-on with their own data as analysis gets offloaded to students, post-docs and statisticians. I'm fighting tooth and nail to avoid becoming one and its still happening. In this context though, it really is necessary. I say all this just to make the point that for the really high-end stuff you really shouldn't focus too much on learning to implement the analysis. Focus on learning what it can do, where to find people who know how to implement it and how to talk to them and strategize about it because this is definitely not like changing the variable names in some SPSS syntax.

I feel like this is really making the case for interdisciplinary work as discussions like these always makes me wonder what separates a psychologist from a computer scientist.

 

[WisNeuro]

I feel like this is really making the case for interdisciplinary work as discussions like these always makes me wonder what separates a psychologist from a computer scientist.

For some, definitely. But, there is pretty wide variability in stats/research knowledge among psychologists, particularly if you include diploma millers. It's one of the reasons you can do well in forensic/IME work if you have a good background as you can absolutely dismantle another provider's reports when they obviously have no handle of the statistics in question.

 

[R. Matey]

For some, definitely. But, there is pretty wide variability in stats/research knowledge among psychologists, particularly if you include diploma millers. It's one of the reasons you can do well in forensic/IME work if you have a good background as you can absolutely dismantle another provider's reports when they obviously have no handle of the statistics in question.

Oh for sure. I was talking about those on the higher end of that continuum, like the schmuck psychology graduate students @Ollie123 was referring to. I am well aware it goes downhill from there. But, where do you draw the line between a stats competent psychologist and a computer scientist? My understanding is that it's pretty rare to find someone who is a solid programmer, statistician, and theoretician. Add clinical practice on top of that and it would be difficult to do all of that well in a training program.

 

[Sanman]

As a stepping stone to a doctoral program, yeah, doesn't really help much. But, as a standalone fallback degree, it has value. Plenty of data science jobs which require the degree.

Yup, go get the degree and then decide if you want the pay cut associated with a PhD once the tech companies come calling.

 

[WisNeuro]

Oh for sure. I was talking about those on the higher end of that continuum, like the schmuck psychology graduate students @Ollie123 was referring to. I am well aware it goes downhill from there. But, where do you draw the line between a stats competent psychologist and a computer scientist? My understanding is that it's pretty rare to find someone who is a solid programmer, statistician, and theoretician. Add clinical practice on top of that and it would be difficult to do all of that well in a training program.

Just find the clinical psych people doing fMRI/eeg work. Most of us had to learn a good deal of programming to get things done

 

[Ollie123]

I feel like this is really making the case for interdisciplinary work as discussions like these always makes me wonder what separates a psychologist from a computer scientist.

100%. I've made the point here before (and gotten some flack for it) that disciplinary boundaries are rapidly becoming a cute historical relic. At least on the research front. I see a lot of assumptions on this board that one should go into psychology if they want to study XYZ (or the converse...that psychologists DON'T study XYZ) when I personally know many exceptions to the rule, many of whom are successfully competing for NIH grants on said topic.

At the same time....there are and there aren't disciplinary boundaries. Having done some of this work, it is really easy to see how blind implementation of models gets you in trouble. Many times the folks working with the data have no content knowledge. Not "Oh they don't fully understand XYZ theory" no content knowledge. Like, they literally don't have even a vague understanding of what the variables they are examining mean, let alone much of the nuance of working with this data. I think ML is the wave of the future and will transform our field, but I also have very strong suspicions even more of it is junk science than we see in the traditional literature. Its very, very, very easy to get positive findings with some data science techniques but they may not mean what you think they mean and probably only get published because peer review isn't designed to capture many of the issues that now come into play. Pile on that as soon as you throw some formulas and fancy charts in that your average clinical journal reviewer's eyes will glaze over and that many of the problems relate to raw data and have no clearly defined solution and you have a big problem.

Obvious and easily understood example is the amount of research I see relying on EHRs with minimal data cleaning. I think everyone who has used one knows records are frequently not up to date, could give countless examples of things they had to put in the chart that aren't "quite" accurate but the best they could fit for some required item, things that technically should be updated but aren't for billing reasons (e.g. I rarely see diagnoses removed/changed following an initial eval even if remitted). Yet these things get blindly thrown into a blackbox model with minimal cleaning and it "predicts" with 75% accuracy who will get better. Well.....maybe. Or it just knows the folks in the phobia clinic improve and the ones in the SMI clinic don't, or that clinicians in clinic XYZ were told to remove a diagnosis in their discharge paperwork and others weren't, etc. No earthly way those issues would get picked up on peer review.

 

[R. Matey]

At the same time....there are and there aren't disciplinary boundaries. Having done some of this work, it is really easy to see how blind implementation of models gets you in trouble. Many times the folks working with the data have no content knowledge. Not "Oh they don't fully understand XYZ theory" no content knowledge. Like, they literally don't have even a vague understanding of what the variables they are examining mean, let alone much of the nuance of working with this data.

Right! Exactly! I was thinking blind implementation of modeling is how you get Cambridge Analytica claiming they measured IQ with Facebook data. They didn't.

I think ML is the wave of the future and will transform our field, but I also have very strong suspicions even more of it is junk science than we see in the traditional literature. Its very, very, very easy to get positive findings with some data science techniques but they may not mean what you think they mean and probably only get published because peer review isn't designed to capture many of the issues that now come into play. Pile on that as soon as you throw some formulas and fancy charts in that your average clinical journal reviewer's eyes will glaze over and that many of the problems relate to raw data and have no clearly defined solution and you have a big problem.

And also what is Machine Learning is an open question. Correct me if I'm wrong, but some of the simpler models are indistinguishable from traditional methods. For instance, I've heard ROCs being referred to as Machine Learning when they have been in around in psychometric research for decades. I sometimes worry that ML is going the way of cognitive neuroscience in the early 2000 - 2010s--where the science couldn't keep up with the enthusiasm.

 

[Sanman]

100%. I've made the point here before (and gotten some flack for it) that disciplinary boundaries are rapidly becoming a cute historical relic. At least on the research front. I see a lot of assumptions on this board that one should go into psychology if they want to study XYZ (or the converse...that psychologists DON'T study XYZ) when I personally know many exceptions to the rule, many of whom are successfully competing for NIH grants on said topic.

At the same time....there are and there aren't disciplinary boundaries. Having done some of this work, it is really easy to see how blind implementation of models gets you in trouble. Many times the folks working with the data have no content knowledge. Not "Oh they don't fully understand XYZ theory" no content knowledge. Like, they literally don't have even a vague understanding of what the variables they are examining mean, let alone much of the nuance of working with this data. I think ML is the wave of the future and will transform our field, but I also have very strong suspicions even more of it is junk science than we see in the traditional literature. Its very, very, very easy to get positive findings with some data science techniques but they may not mean what you think they mean and probably only get published because peer review isn't designed to capture many of the issues that now come into play. Pile on that as soon as you throw some formulas and fancy charts in that your average clinical journal reviewer's eyes will glaze over and that many of the problems relate to raw data and have no clearly defined solution and you have a big problem.

Obvious and easily understood example is the amount of research I see relying on EHRs with minimal data cleaning. I think everyone who has used one knows records are frequently not up to date, could give countless examples of things they had to put in the chart that aren't "quite" accurate but the best they could fit for some required item, things that technically should be updated but aren't for billing reasons (e.g. I rarely see diagnoses removed/changed following an initial eval even if remitted). Yet these things get blindly thrown into a blackbox model with minimal cleaning and it "predicts" with 75% accuracy who will get better. Well.....maybe. Or it just knows the folks in the phobia clinic improve and the ones in the SMI clinic don't, or that clinicians in clinic XYZ were told to remove a diagnosis in their discharge paperwork and others weren't, etc. No earthly way those issues would get picked up on peer review.

Absolutely agree, while not working at your level, the foundation of my career (lab work etc) was on the cognitive and neuropsych side. Even the pure psych folks needed someone to program their experiments. While, I was not the greatest programmer, I spoke both psychology and software code. This got my foot in the door at many labs. This was almost 20 years ago. Things have become more integrated in that time, not less.

 

[WisNeuro]

Absolutely agree, while not working at your level, the foundation of my career (lab work etc) was on the cognitive and neuropsych side. Even the pure psych folks needed someone to program their experiments. While, I was not the greatest programmer, I spoke both psychology and software code. This got my foot in the door at many labs. This was almost 20 years ago. Things have become more integrated in that time, not less.

Did it take a long time to make all of the punch cards you needed?

 

[Sanman]

Did it take a long time to make all of the punch cards you needed?

Yeah, but it gave me something to do while my dial up modem accessed AOL. Remember that?!?!

 

[WisNeuro]

Yeah, but it gave me something to do while my dial up modem accessed AOL. Remember that?!?!

Oh yeah, my connection in HS was the ole 14.4, which we later upgraded to 28.8, which made downloading por...I mean music from Napster that much faster.

 

[Dazen]

Lastly, recognize this is going to be tough to REALLY dig in on graduate school and you probably won't get to do anything truly crazy/innovative for a thesis/dissertation and that's OK. Tradition in psychology - for understandable reasons - is you collect your data and run your own analyses. These things require an obscene amount of data that is often untenable for a student project. I don't know your background, but chances of you actually being able to do much beyond the simplest analysis yourself is relatively slim. Any schmuck psychology grad student can run a factor analysis or even a fairly complex random effects model. Good ones can implement fancy latent growth models or computational cognition models. Could you figure out how to run a lasso regression? Sure. Could you implement a multilayer perceptron classifier embedded inside a deep neural network that scaffolds off Inception while folding in a Gaussian Process model? I've been coding since I was 10, darn near went into computer science instead of psychology, was the stats geek in grad school and I might retire before I got that done successfully. This kind of stuff we have had multiple PhD-level engineers working for months to implement analysis for a single research question. So even if you "could" do it yourself....you won't because it isn't an efficient use of time. I think it is very much a bad thing that virtually all senior psychologists move away from being hands-on with their own data as analysis gets offloaded to students, post-docs and statisticians. I'm fighting tooth and nail to avoid becoming one and its still happening. In this context though, it really is necessary. I say all this just to make the point that for the really high-end stuff you really shouldn't focus too much on learning to implement the analysis. Focus on learning what it can do, where to find people who know how to implement it and how to talk to them and strategize about it because this is definitely not like changing the variable names in some SPSS syntax.

This is so helpful to remember. I know that the earlier part of the PhD is more course-heavy, but I've already been frustrated by how much time I need to put into coursework/TA responsibilities instead of being able to focus more on research (and I generally work quickly, and have put comparatively less time in than the rest of my cohort). I'm fortunate that my advisor allows me to be very independent in doing my own research, running analyses on previously collected data, and spending a little bit of time per week doing "fun math!" but it's still surprising how quickly all the hours in the week get used up. I'm likely (similar to most grad students/academics) to overcommit myself and get excited by many different ideas to the detriment of having enough time for any individual one. It's hard to prioritize exactly what to focus on and I know it's not going to get better, but want to maintain some connection to what got me excited about grad school in the first place, even while remaining open to new passions.

To clarify my background in a broad sense, I majored in engineering in college (concentrated basically in applied math at a very theoretically-oriented program), almost double majored in computer science, and still sometimes unwind by watching math lectures on MIT Opencourseware, so not the most traditional background for psychology. I know I have an incredible amount to learn, but I think I have a decent base knowledge and natural ability.

 

[Sanman]

This is so helpful to remember. I know that the earlier part of the PhD is more course-heavy, but I've already been frustrated by how much time I need to put into coursework/TA responsibilities instead of being able to focus more on research (and I generally work quickly, and have put comparatively less time in than the rest of my cohort). I'm fortunate that my advisor allows me to be very independent in doing my own research, running analyses on previously collected data, and spending a little bit of time per week doing "fun math!" but it's still surprising how quickly all the hours in the week get used up. I'm likely (similar to most grad students/academics) to overcommit myself and get excited by many different ideas to the detriment of having enough time for any individual one. It's hard to prioritize exactly what to focus on and I know it's not going to get better, but want to maintain some connection to what got me excited about grad school in the first place, even while remaining open to new passions.

To clarify my background in a broad sense, I majored in engineering in college (concentrated basically in applied math at a very theoretically-oriented program), almost double majored in computer science, and still sometimes unwind by watching math lectures on MIT Opencourseware, so not the most traditional background for psychology. I know I have an incredible amount to learn, but I think I have a decent base knowledge and natural ability.

 

[Dazen]

But I bet I could out bench you, bro

 

[Sanman]

But I bet I could out bench you, bro

Now, I bet you could as well. Rotator cuff injury and 18 mths away from weight training. In grad school, I was benching 315lbs with a 405lb deadlift.

 

[WisNeuro]

Now, I bet you could as well. Rotator cuff injury and 18 mths away from weight training. In grad school, I was benching 315lbs with a 405lb deadlift.

Impressive, more than I was benching in my prime before shoulder surgery. Still, there is an SDN reg here who I know can outbench the entire forum.

 

[Sanman]

Impressive, more than I was benching in my prime before shoulder surgery. Still, there is an SDN reg here who I know can outbench the entire forum.

I'm pretty sure I know who and I'm sure I would lose.

 

[WisNeuro]

I'm pretty sure I know who and I'm sure I would lose.

Seriously, I had to do a double take and recalculate ow much was actually on the rack before believing it. A truly humbling experience.

Actually, I think we have at least a couple power lifters here, I'd be interested in who'd prevail on certain lifts. We may be thinking of a different posters.

 

[Sanman]

Seriously, I had to do a double take and recalculate ow much was actually on the rack before believing it. A truly humbling experience.

Actually, I think we have at least a couple power lifters here, I'd be interested in who'd prevail on certain lifts. We may be thinking of a different posters.

Had a grad school professor who was a powerlifter and could out lift me. Seems to be a thing. I am far past my prime though. Been focusing on lower weight and higher reps due to injuries. Just trying to stay healthy and uninjured.

 

[WisNeuro]

Had a grad school professor who was a powerlifter and could out lift me. Seems to be a thing. I am far past my prime though. Been focusing on lower weight and higher reps due to injuries. Just trying to stay healthy and uninjured.

Likewise, just looking for general fitness over strength. Just want to maintain my volleyball/softball rec league game. Though, a buddy of mine who used to do MMA has been wanting to do more training and sparring, so have been getting more into that, lower key than full-go though.

 

[Sanman]

Likewise, just looking for general fitness over strength. Just want to maintain my volleyball/softball rec league game. Though, a buddy of mine who used to do MMA has been wanting to do more training and sparring, so have been getting more into that, lower key than full-go though.

Been trying to get back into boxing/ kickboxing recently. Great cardio without a ton of equipment.

 

[Dazen]

Been trying to get back into boxing/ kickboxing recently. Great cardio without a ton of equipment

Likewise, just looking for general fitness over strength. Just want to maintain my volleyball/softball rec league game. Though, a buddy of mine who used to do MMA has been wanting to do more training and sparring, so have been getting more into that, lower key than full-go though.

I used to do muay thai (never competitively) and it was a fantastic workout and great stress reliever. I've never actually been that into powerlifting. I do more ultra-distance hiking (up to 50 miles in a day) and functional fitness (kettlebells, sandbag, high intensity bodyweight stuff), plus gotten pretty into rucking lately. I really want to do Goruck Selection next year, but not expecting any success haha.

 

[WisNeuro]

I used to do muay thai (never competitively) and it was a fantastic workout and great stress reliever. I've never actually been that into powerlifting. I do more ultra-distance hiking (up to 50 miles in a day) and functional fitness (kettlebells, sandbag, high intensity bodyweight stuff), plus gotten pretty into rucking lately. I really want to do Goruck Selection next year, but not expecting any success haha.

Sounds fun. I do a form of rucking that uses a toddler. In that when we go for a hike he eventually refuses to go any more and then I have to carry him.

 

[Ollie123]

This is so helpful to remember. I know that the earlier part of the PhD is more course-heavy, but I've already been frustrated by how much time I need to put into coursework/TA responsibilities instead of being able to focus more on research (and I generally work quickly, and have put comparatively less time in than the rest of my cohort). I'm fortunate that my advisor allows me to be very independent in doing my own research, running analyses on previously collected data, and spending a little bit of time per week doing "fun math!" but it's still surprising how quickly all the hours in the week get used up. I'm likely (similar to most grad students/academics) to overcommit myself and get excited by many different ideas to the detriment of having enough time for any individual one. It's hard to prioritize exactly what to focus on and I know it's not going to get better, but want to maintain some connection to what got me excited about grad school in the first place, even while remaining open to new passions.

To clarify my background in a broad sense, I majored in engineering in college (concentrated basically in applied math at a very theoretically-oriented program), almost double majored in computer science, and still sometimes unwind by watching math lectures on MIT Opencourseware, so not the most traditional background for psychology. I know I have an incredible amount to learn, but I think I have a decent base knowledge and natural ability.

Normal part of the grad school experience and one I imagine most of us on a research path felt to some extent. I suspect you are like me and lean at least slightly more towards basic science where I think it is felt even more strongly - if ones primary interest is therapy RCTs, a lot of the clinical coursework is more relevant than if you are interested in the effects of early childhood trauma on white matter integrity. Time will free up later on for you to do more stuff like this and potentially even take electives on relevant topics if you can make an appropriate case for it to your department.

No idea what your research interests are, but shoot me a PM in 4ish years if you are looking for a post-doc.