In Conclusion

• Published in: Statistics in Corpus Linguistics Research Vol. 1; pp. 314 - 316
• Main Author: Wallis, Sean
• Format: Book Chapter
• Language: English
• Published: Routledge 2021
• Edition: 1
• Subjects: Experimental Design Weaknesses; OLS Regression; Log Likelihood Tests; Confidence Interval Estimates; Random Sentences; Model Random Sampling; Typical Error Level; Finite Population Correction; Wald Interval; Linguistic Evaluation; Vice Versa; Corpus Linguistics Research; Wilson Interval; Solve Research Problems; Domain Specific Theorising; Interval Assessment; Multi-variate Analysis; Comparing Test Results; Cochran's Rule; Yates's Methods; GB; Wilson Score; Wilson Score Intervals; Plotting Data
• ISBN: 9781138589377; 9781138589384; 1138589381; 1138589373
• DOI: 10.4324/9780429491696-25

For a researcher, the entire point of a statistical method is to understand the implications of their data. Research concerns the structure of the physical world: in linguistics, the structure of language. The traditional approach to teaching experimental statistics revolves almost exclusively around significance testing. Distributions are skipped over rapidly, intervals are barely mentioned. In this chapter, the author shows that statistical methods may be used to support the laborious task of manually reviewing sampled data - a necessary and often overlooked step in corpus research. The author uses statistical methods to improve sampling precision and model random sampling. But statistical methods do not turn a poor experimental design into a good one. All experiments represent a compromise. It may be possible for another researcher to improve on our experiment: for example, to employ a controlled laboratory experiment to review the results of a corpus one.