Sometimes you read research that aligns beautifully with your vision. And sometimes it clashes with what you already knew, or thought you knew. For example, because it seems to contradict previous research you have read. Or worse, because it is not necessarily in line with… your own research. But that is precisely why I felt obliged to present this research on the relationship between students and their teacher! Recent studies highlight how essential the teacher-student relationship is for academic achievement.
There is a considerable amount of research on this topic. A warm, positive relationship is associated with various positive outcomes, including engagement and academic performance. There are also longitudinal studies suggesting that this relationship may be reciprocal. This means that a better relationship with the teacher predicts better performance later on, while better performance, in turn, can also predict a better relationship with the teacher. I myself conducted research into the role of authenticity in this.
But then there is this new study by Daniel Hajovsky and colleagues in the Journal of School Psychology. This research poses a rather fundamental question: is that effect real, or are we seeing it partly because we are using the wrong statistical lens? Are you triggered yet? I am.
An old discussion with a new analysis method
The researchers focused on two aspects of the student-teacher relationship: closeness, roughly warmth and closeness in the relationship, and conflict . These were measured using the widely used Student-Teacher Relationship Scale. In their analyses, the researchers also examined math performance. This is not new and aligns with previous research. The twist lies in the way they analyse the data.
Much earlier longitudinal research used so-called cross-lagged panel models . With this methodology, for example, it is possible to examine whether the relationship with the teacher at one point in time predicts performance at another, while accounting for prior performance. Several studies have, in fact, provided evidence of effects in both directions with such a model. Among other sources, the researchers cite research conducted in the Netherlands by Zee et al., involving 1,267 pupils, that demonstrates, for example, reciprocal influences between the pupil-teacher relationship and reading and math performance.
However, Hajovsky et al. raise an issue regarding this approach. Dynamic models fail to adequately distinguish between inter-individual and intra-individual differences, and thus a correlation that appears to be a dynamic effect may actually be partly due to something else. I have to admit, this is complex stuff, and here I largely have to rely on the authors’ explanation.
As an alternative statistical approach, the authors of this new study use latent change score models. The interesting question then becomes not only: does a good relationship predict later performance? They look more specifically at whether the quality of the relationship predicts a subsequent change in performance.
847 children, followed throughout primary school
The researchers used data from the American Study of Early Child Care and Youth Development for their analysis. The original sample consisted of 1,364 children who were followed long-term starting in the 1990s. For these analyses, 847 children with sufficient measurements ultimately remained. The relationship with the teacher was measured annually from the first to the sixth grade. Mathematics performance was measured in the first, third, and fifth grades using a standardised test.
And now it turns out that the researchers, using their analytical method, find no evidence that differences in closeness to or conflict with the teacher predict later changes in math performance. Furthermore, they do not find the reverse either. Performance also does not predict a subsequent change in the relationship with the teacher.
And it is not the first time
While reading this study, I discovered once again that research can still escape my notice. This study turns out to be a replication. In 2025, Hajovsky and Chesnut had already applied the same analysis method to another large American longitudinal dataset with children from a different generation. Even then, they found no evidence of that mutual influence. Variation in the student-teacher relationship did not predict later changes in reading or math performance, and differences in performance did not predict a change in the relationship either.
Because these results did indeed conflict with a substantial body of existing research literature, the researchers decided to repeat their analysis using an independent dataset. And once again, they arrive at broadly the same result. That is immediately one of the study’s strong points: the authors explicitly attempt to replicate a finding that contradicts what is often assumed in their field of research.
And now it is getting uncomfortable for me.
For years, I argued in my own research that a good student-teacher relationship had almost as much of a learning effect as good feedback (referring to Hattie, 2009, among others). May I now bury this argument for good? I don’t think so. This new study is, however, a useful warning to be very precise about what such results might mean. When we see in datasets that students with a better relationship with their teacher perform better on average, we see a correlation. That is still different from demonstrating that improving that relationship subsequently improves their performance. I am now grappling with the question myself of whether I have also fallen into the causal trap.
Moreover, this study shows that the choice of statistical model can help determine which narrative emerges from the same type of longitudinal data. The researchers therefore compare different analytical methods and explicitly ask whether commonly used models may have steered us too easily towards a reciprocal effect in the past.
Don’t throw the relationship with your students away just yet.
The authors begin their article by citing research showing that supportive relationships are important for, among other things, school adjustment and engagement. Relationships can therefore matter for many different reasons, even if we cannot demonstrate that improving them directly leads to higher math scores. So don’t read this study as evidence that relationships don’t matter. I didn’t.