Nathan McJames. Image: John Reilly
‘I hope to see my research inform policy decisions that can directly improve the lives of students and teachers,’ says Nathan McJames.
With a BSc in science with education and an MSc in data science and analytics, Nathan McJames planned to become a maths and physics secondary school teacher. However, he got the research bug and when the opportunity arose to combine his love of statistics and teaching, he jumped at it. He’s currently finishing up a PhD in the Hamilton Institute at Maynooth University, where he’s developing new methods to analyse large education datasets to uncover insights for policymakers and educators.
One project he worked on looked at factors influencing teacher job satisfaction. It was found that continuing professional development (CPD) has a positive impact on teacher satisfaction, while part-time contracts tend to have a negative effect. “These insights can help mitigate teacher shortages by informing policies that encourage teachers to remain in the classroom and attract new teachers to join the profession,” McJames said.
Here, he tells us a bit more about the new methods he’s developing in his research.
Tell us about your current research.
A common challenge when working with observational data is that it can be very difficult to make claims of a causal nature. My research during my PhD has focused on the development of new causal inference machine learning methods, that in the right conditions, can accurately estimate causal effects from observational data. I have also applied these methods to large-scale educational datasets to tackle research questions with important policy implications.
Much of my research has focused on taking existing causal inference methods and extending them to make them applicable in situations they would not normally be well suited to. For example, I have developed extensions that allow existing methods to be applied to not just one, but multiple outcome variables, or allow them to estimate the effects of multiple factors simultaneously.
The most recent project I worked on involved developing a longitudinal extension of a causal inference method called Bayesian Causal Forests. This was motivated by a longitudinal dataset called the High School Longitudinal Study of 2009, which tracked the mathematics achievement of high-school students in the US over time.
Using this new method, we were able to investigate how the students’ achievement developed over time, and how this development was affected by whether or not the students participated in part-time work during the school year. Key findings from the study revealed that students who started with higher achievement were more likely to experience significant gains, while those with lower initial achievement showed less progress, suggesting a widening achievement gap. Additionally, we found that on average, part-time work was associated with a modest but clear reduction in achievement growth.
These findings could have several important implications. The finding of a widening achievement gap, for example, highlights the need for targeted interventions to support low-achieving students and help bridge this divide. Additionally, the negative impact of part-time work on academic performance could lead to a re-evaluation of policy on working hours for young students, and how students balance work and study.
In your opinion, why is your research important?
Traditional randomised controlled trials (RCTs), while the gold standard, are often prohibitively expensive or ethically impossible to perform. Therefore, with large-scale observational datasets from education and other fields becoming increasingly common, it is imperative that researchers have the right tools available to study and make meaningful inferences from these datasets. That’s why the methods I work with and help to develop are so important.
I’d like to see my research having a meaningful impact in two key areas. First, I hope that the methods I’ve developed will be adopted by researchers not only in education but also in other fields, such as healthcare, economics and social sciences, where observational data is prevalent.
Second, I also hope to see the insights gained from my research on education datasets inform policy decisions that can directly improve the lives of students, teachers and schools. By providing evidence-based recommendations, I aim to contribute to policies that improve educational outcomes and address challenges such as teacher shortages, student achievement gaps and equitable access to resources.
Ultimately, I hope that my work can help shape policy documents and influence decisions that lead to tangible improvements in education systems worldwide.
What inspired you to become a researcher?
I don’t think I can pinpoint any specific memories that inspired me, or the specific moment I decided to pursue my research, but I’ve always enjoyed being able to dive into problems that really interest me, and pursuing a PhD gave me the perfect opportunity to do just that.
Before my PhD I actually trained as a teacher and was on the path to becoming a secondary school teacher. I really enjoyed this experience, but ultimately decided to follow my other passion of statistics and data science instead. Then, when I discovered the PhD opportunities in statistics at the Hamilton Institute, I saw a unique chance to combine these two passions – using data science and statistical methods to address significant challenges in the education sector. This blend of interests has shaped my research journey ever since.
What are some of the biggest challenges or misconceptions you face as a researcher in your field?
I think one common misconception about observational data is that it is impossible to answer causal questions with it. In fact, this is something I thought too before I started studying and researching in this area. While it’s true that answering causal questions with observational data can be difficult and is not always possible, the right tools, strategies and datasets can enable researchers to gain robust and meaningful insights, even when randomised trials are not possible.
Specifically, the type of methods I have worked with, when applied under the right conditions, can leverage the advanced predictive capabilities of modern statistical and machine learning methods to control for the effects of confounding variables that might otherwise skew the results of an analysis, or trick us into thinking there is a causal effect present when really there isn’t or vice versa.
It’s always important to remember the limitations and assumptions of these methods, however, and be open and honest about these when sharing findings.
Do you think public engagement with science and data has changed in recent years?
I think this is definitely true – the Covid-19 pandemic highlighted the huge role that data and our analysis of it plays in making decisions that affect the lives of all of us. This is also true of the field that I research in – education – because large decisions that can affect thousands of students and teachers are often based on the types of datasets that I work with.
Because the data I work with is so important and the decisions that arise from analyses can be too, I feel very lucky to be able to research in this area and help contribute to answering important research questions that can help to benefit so many people. However, it also brings a sense of responsibility and pressure. I feel a strong obligation to ensure that my research and analyses are performed as well as possible, which is why I am committed to developing new and powerful statistical tools to model and interpret this data with the highest degree of precision.
How do you encourage engagement with your work?
One of the things that I’ve enjoyed most about my education research is that so many people have an interest in this area. Whether you’re a student, teacher, parent or simply thinking back to your own school days, I think everyone can find something to relate to in my research.
Recently, a project I worked on which examined the effect of homework on student achievement was discussed in the news and on the radio, so it was great to see this part of my research reaching a wider audience and being discussed by members of the public.
In addition to reaching wider audiences, I make all the code from my projects open source, enabling others to use and adapt it for similar research. I’ve also been working on developing one of the causal inference methods I created into an R package, which allows fellow researchers to easily incorporate it into their own work. My goal is to continue building similar tools and R packages for the methods I develop, fostering ongoing engagement and collaboration within the research community.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
