This is part of a series of analyses of the relationships between middle-distance events. The 800m is the focus of this study, but the data offer a lot to learn about each discipline. The main portion of this analysis surrounds the impact of the 400m and 1500m performances on 800m race performances at different levels of competition. Other products will include prediction of 800m times based off of 400m and/or 1500m times.
The Influence of the 400m and 1500m on 800m Race Performances
Background Information
The data for this project come from the MileSplit.com database and the TFRRS.org database. Indoor and outdoor performances were recorded across the 400m, 800m, and 1500m. 1600m and mile times were converted to 1500m times. Observations in the dataset contained a single athlete who, in a single season, competed in all three events, and whose season's best ranked in up to the top 2000 performances listed in the TFRRS database, or the top 1000 performances in the MileSplit database.
The data were obtained using custom webscrapers I built that are available on my GitHub. The data were lightly cleaned to remove obvious outliers, but otherwise were not scaled, centered, or transformed. I split the full dataset into two smaller sets according to sex. The subtitles of each plot indicate which sex(es) the data were modeled on, whether or not it was the cleaned (no notation) or uncleaned (raw) dataset, and the size of the dataset.
Conclusion
So far, the results from this analysis show that although the 400m has a larger effect size than the 1500m, its influence on the 800m is relatively constant. Conversely, an 800m runner's ability to run a faster 1500m will definitively lead to faster 800m performances. In non-technical terms, 800m runners need the speed required to run a fast 400m. Without that speed, runners won't run fast 800m times. However, improving individual 400m performance won't necessarily change 800m performance. Instead, 800m runners that incorporate high intensity aerobic fitness, like that required to run a fast 1500m, will see stronger increases in 800m performance as long as they maintain their speed.
Notes
The dotted lines in the visualizations below represent 95% confidence intervals for the OLS parameter estimates and the quantile regression parameter estimates, and the solid lines represent the parameter estimates. When the confidence intervals overlap, we cannot confidently say that the respective OLS parameter significantly differs from the respective quantile parameter.
Furthermore, this is a preliminary analysis. The two data sources have differences in in distributional variance, and that needs to be investigated before any final conclusions are made.