A Systematic Review and Meta-analysis of Text Entry Studies

Text entry is a notably standardized research field in human-computer interaction, with established benchmarks and methodologies enabling rigorous comparisons across studies. Nevertheless, meta research in text-entry is scarce. This meta-analysis summarizes findings and effects from text entry experiments published from 1990 to 2024. Our records show that most text-entry experiments feature a baseline and an experimental UI, and that they mostly show that the experimental UI is superior ($g = 1.68$), with regards to text entry speed. We found that earlier text entry research focused mostly on the development of novel techniques, whereas recent text entry research, to a larger extent, adapts existing input methods to new devices, such as smartwatches or virtual reality headsets. We also find that text entry research often lacks statistical power ($M=0.66$, $SD=0.37$), relies on small sample sizes ($Mdn=12$), and is mostly conducted with within-subjects designs (84. Subjective evaluations of text entry systems beyond objective performance are rare. Our analysis found evidence for systematic publication bias in text entry research, as indicated by funnel plot asymmetry and a significant weight-function model adjustment. This underlines the research field's competitive culture of publishing research only if entry speed is beaten in comparison to some baseline.