The State of the Dog Training Research
I've been having some interesting conversations on here about the state of the research on dog training, and I’ve been genuinely surprised by how confident some interpretations of the literature seem. I’ve been reading it closely, and I've come away thinking that the evidence is considerably less definitive than online discussions often suggest.
There are a number of studies comparing reward-based and aversive-based training methods. However, the body of research remains relatively small and is characterized by significant methodological limitations, making it difficult to draw broad causal conclusions or generalize the findings.
Most research compares training philosophies while giving comparatively little attention to other variables that may have equal or greater influence on canine welfare, including trainer competence, owner behavior, genetics, temperament, environmental management, and consistency of implementation. The implicit assumption in much of the discussion is that the training method used is a primary determinant of long-term welfare, but that assumption has not yet been demonstrated. One recent study, for example, has found that owner intervention, rather than training method, was actually the strongest predictor of improvement in canine reactivity (Loughridge, 2025).
The study most frequently cited by proponents of force-free training (Vieira de Castro et al., 2020) suggests that dogs trained using aversive methods display more stress-related behaviors. However, the study has several important limitations, the most significant of which is that it was not a randomized controlled trial. Dogs were not randomly assigned to reward-based or aversive-based groups, creating a substantial risk of selection bias. It is therefore hard to know whether the observed differences resulted from the training methods themselves or from pre-existing differences among the dogs, owners, or trainers.
In addition, the training protocols were not standardized. Numerous variables differed simultaneously between the groups, including trainer skill, timing, reinforcement rate, class structure, handling ability, owner coaching, and training environment. The study also grouped a wide range of techniques, including leash pops, physical manipulation, body blocking, and yelling, under a single label, "aversive," making it difficult to determine which technique was responsible for the observed effects. The sample was relatively small and drawn from a single country, limiting generalizability across different training cultures, breed distributions, and owner demographics. Finally, the study was not designed to determine whether the short-term indicators of stress it identified translated into poorer long-term welfare or behavioral outcomes, nor did it examine dogs with severe behavioral problems. While the findings are certainly relevant, I don't think they should be generalized beyond the scope of the study.
More recent research has included a randomized controlled trial suggesting that a structured e-collar protocol was more effective than a reward-based protocol, with no significant differences in measured stress (Johnson & Wynne, 2025). However, this study has also received substantial criticism, with critics noting that it compared a relatively well-developed e-collar protocol with a comparatively weak reward-based protocol rather than the strongest versions of each approach. The e-collar group also received an additional training session, complicating the interpretation of the results. Other limitations include the small sample size, the narrow training task, the relatively short follow-up period, and welfare measures that may not have been sensitive enough to detect more subtle emotional effects.
There are also several notable gaps in the literature. To my knowledge, no published studies have examined the long-term welfare outcomes of dogs trained using different methods over periods of years rather than weeks or months. Many studies reporting short-term welfare measures are frequently discussed as though they establish long-term consequences, but those outcomes have not actually been studied. Likewise, there appear to be no studies comparing the effects of high-quality versus poor-quality training regardless of philosophy, even though trainer competence is likely to have a substantial impact on both welfare and behavioral outcomes. Nor have studies examined whether different training approaches differ in how difficult they are for trainers or owners to implement correctly.
One additional question that I think deserves study is why the increasing popularity of reward-based training has coincided with changes in the prevalence of behavioral problems such as fear, anxiety, aggression, or reactivity. We know that reward-based approaches have become increasingly popular over the past several decades (Johnson & Wynne, 2024; Woodward et al., 2021; DeLeeuw & Williams, 2026), and we also know that behavioral problems remain common in companion dogs (Beaver, 2025; Beaver, 2026). No study has directly investigated whether a causal relationship exists or whether changes in training preferences have had any measurable association with population-level behavior trends. These things may be completely unrelated, but it simply strikes me as an obvious question that has yet to be examined.
My takeaway isn't that one philosophy has been proven superior to the other. It's almost the opposite. I think the current literature provides useful evidence, but it remains preliminary. Given the methodological limitations and the substantial gaps that still exist, I don't think the science yet supports the degree of certainty that often appears in online discussions.
What am I missing? I'd be interested to hear whether others who have read the literature have reached a different conclusion, particularly if there are long-term welfare studies or well-controlled trials that I've overlooked.