The role of statistical analysis in modern nursing research

Abstract

The task of analyzing research data has changed greatly over the past 30 years. Performing complex statistical calculations by hand is now obsolete. Statistical software packages allow statisticians to conduct data analysis much faster and with better accuracy. In addition, the ease of use of most statistical software applications provides non-statisticians with the ability to conduct their own data analysis. Dramatic improvements in computing technology, coupled with those in statistical software, have also provided researchers with access to a wider array of statistical methods. Bayesian analysis, with its computationally-intensive methods, is finding increased application in health science research. The burgeoning field of data science, with its promise of finding patterns in “big data”, is another option for those researchers wanting to explore giant data sets with millions of observations. With all of this statistical power now at their command, nursing researchers may view the role of statistical analysis differently. Should I alter my aims and hypotheses to include more sophisticated statistical methods? Won't reviewers of grant proposals and manuscripts be expecting the use of more and more advanced statistical methods? Nursing researchers may fear their grant proposal won't be funded or their manuscript won't be published if their statistical methods are not seen as “complicated” enough. However, the basic research process, and the role that statistical analysis plays in that, has not changed. All research studies should be based on questions or hypotheses. Any statistical methods used for a study should be based on those hypotheses, taking into account specific characteristics of the variables and the design of the study. A researcher should never let the statistical methods (or the statistician) dictate their hypotheses. The formulation of a hypothesis should be grounded in the nurse researcher's clinical knowledge. The choice of the statistical method follows logically from that: it is simply the appropriate tool to evaluate the hypothesis. All of which underscores the importance of creating a data analysis plan during the development of a research study (Harrington et al., 2019). The data analysis plan provides a detailed description of all statistical methods used for a study. It should be organized according to the aims and hypotheses of the study to provide a blueprint for how the analysis will be conducted. Although the statistician is typically responsible for determining the data analysis plan, it should always be done in consultation with the researcher. Once the data are collected and analysis has begun, it is important to not deviate from the analysis plan to avoid running too many statistical tests and finding differences or associations that don't exist in the population (Type I error). Maintaining the primacy of the nurse researcher in determining the hypotheses should also be extended to the interpretation of the results. Viewing the results of data analysis as only statistically significant or not removes the researcher from the interpretation. This is partly why the American Statistical Association adopted the policy of no longer using the term “statistically significant” (Wasserstein, Schirm, & Lazar, 2019). The interpretation of study results is best viewed from the perspective of clinical significance, or does a difference or association found in the sample have a practical effect on patient care? Nursing researchers may find it peculiar that a statistician is advocating for restraint when using statistical analysis. But it should always be that way. It is the nurse researcher's hypothesis that is of primary importance. The role of statistical analysis is to evaluate that hypothesis: does the evidence from the sampled data support the researcher's hypothesis? And that role has not changed, even with all of the modern advancements to how we perform statistical analysis.