A reflection of ‘Training Strategies to Improve Muscle Power: Is Olympic-style Weightlifting Relevant?’ – a need for bayesian statistical analyses in sport science research.

Helland et al. (2017) compared 8 weeks of Olympic lifting, strength training and motorized strength and power training (isokinetic and isotonic) in 39 athletes (20±3years old) randomized into respective groups. All groups participated in 2-3 training sessions per week for an 8-week study duration. Training volume was equivalent between groups. Measures of countermovement jump (CMJ), squat jump (SJ), drop jump (DJ), loaded countermovement jump(LCMJ), 30-metre sprint and 1RM squat were taken at baseline and post-intervention.

Results demonstrated small improvements in 1RM squat, CMJ and SJ in the Olympic lifting group. Additionally, small reductions in DJ performance were noted in the Olympic lifting group. The strength training and motorized strength and power training groups improved on all variables and demonstrated a larger magnitude of change compared to the Olympic lifting group.

Results demonstrate evidence that Olympic lifting may be less effective than strength training and motorized strength and power training at improving lower body power, strength and reactive strength. This has important implications for strength and conditioning coaches who prescribe Olympic lifts to enhance physical performance and necessitates further research to confirm the findings of this study.

Methodologically, the study was well organised and appropriate training interventions were prescribed. The small number of participants will have had negative effects on statistical power (n=13 for each intervention group). Future research should consider recruiting more participants and reducing the number of comparison groups. Alternatively, given the practical constraints of recruiting large numbers of appropriate participants researchers should consider utilising Bayesian estimates. Bayesian analyses have the advantage of utilising prior information as a method of informing and narrowing the highest density interval, thus enabling a more precise estimation.

The main difference of Bayesian analyses to frequentist statistical analysis is the approach to defining a problem. Frequentist approaches aim to measure the probability of the null hypothesis being true given the data collected. This involves formulating a hypothesis, collecting data and using null hypothesis testing (NHT) to confirm or reject the null hypothesis. Bayesian analysis uses prior probability (either from previous research or pilot testing) which is updated to a posterior probability given new and relevant data. The Bayesian approach results in an estimation of the degree of uncertainty. Compare this to the frequentist p-value which can result in errors in interpretation. For example, a p-value of 0.051 is 0.1% higher than a p-value of 0.05. The frequentist method of interpretation can lead researchers to reject the null hypothesis is the p-value is <0.05, yet accept the null hypothesis if the p-value is 0.051.

Bayesian analysis will have a specified prior, informed by prior research, which informs the precision of the highest density interval (HDI). Data collected and analysed will subsequently enable us to estimate the location and spread of the HDI and enable researchers to estimate the probability (of for example differences in means) and express the confidence that can be placed in the difference in means being present, rather than accepting or rejecting a null hypothesis (e.g. no difference between means). If we accept the null hypothesis on the basis of p=0.075 or p=0.051, the cause may be low statistical power (participant numbers being inadequate) rather than the intervention being ineffective. Thus, we may reject an efficacious intervention due to having low participant numbers. However, using the Bayesian approach, we can use the data of the individual study to inform future similar research by defining prior probabilities that can be used to further narrow the HDI (a method of updating the evidence based on new presented data – a continual accumulation of priors being informed by new data similar to the frequentist meta-analysis). Therefore, the low statistical power of the current study could be avoided in future studies by using a bayesian approach that could inform further studies, with each study narrowing the precision even further.

Utilising the current frequentist analytical approach, further studies are required to replicate these findings before accurate conclusions can be confirmed. However, a tentative conclusion from this study is that Olympic lifting is less effective than strength training at improving measures of power, strength and reactive strength. This may be explained by the individual limiting factors to strength, power and reactive strength development. Specifically, if the group’s rate of force development had accommodated and plateaued, strength training would necessarily be the primary focus of training to raise force production capacity. Following an increase in force production, training could again focus on developing rate of force development to raise the speed at which maximum force can be reached. Thus, individual limiting factors (e.g. force production, rate of force development, musculotendinous stiffness) at any point in time may be influencing the training need. Consequently, this group may have required strength training to further elevate maximum force production before training for increased rate of force development.

 

References

 

Helland, C., Hole, E., Iversen, E., Olsson, M. C., Seynnes, O., Solberg, P. A., & Paulsen, G. (2017). Training Strategies to Improve Muscle Power: Is Olympic-style Weightlifting Relevant? Medicine & Science in Sports & Exercise, 49(4), 736-745.