Abstract:

Literature research conducted in education and agricultural education journals published during a period of 10 years revealed that 98% of the studies used parametric analyses. In general, model assumptions were not tested, and statistical criteria were not followed to apply the parametric approach. The objective of this presentation is to persuade researchers to use the most appropriate statistical analysis for their data. We present a case study in agriculture education where a parametric multiple linear regression could be applied. A survey was designed to find out how Theory of Planned Behavior and Importance-Performance variables were associated to Behavioral Intent concerning landscape water conservation practices. Total sample size was 2,118 completed surveys.

The research hypotheses were the following:
(a) Planned Behavior variables such as Social Norms (SN), Perceived Behavioral Control (PBC), and Attitude had a linear relationship with Behavioral Intent (BI).
(b) Importance-Performance variables such as Importance of clean water for local and large water bodies (ICW), satisfaction with clean water for local and large water bodies (SCW); and Personal Norms (PN) had the potential to challenge the predictive power of the above three on BI.
(c) SN, PBC, A, ICW, SCW, and PN could possibly provide the greatest predictive power on BI.

Although model assumptions were not met, we initially carried out a multiple linear regression (MLR) analysis based on the premise that the results could be reported descriptively if they were double cross-validated successfully. The double cross-validation of the MLR was not successful, and model assumptions were not held even though the sample size was large. A quantile regression model fitted the data well, and we modeled the median of BI. Theory of Planned Behavior and Importance-Performance variables were good predictors of BI, excepting Attitude. We demonstrate that imposing the Central Limit Theorem to use the MLR model is not the correct criterion to apply a parametric approach. We should use double cross-validation. Researchers must rely on statistical criteria to support decisions regarding the use of parametric or non-parametric procedures. The fact that we encountered such a low percentage of non-parametric analyses in an interval of 10 years in disciplines where variables in general are skewed, motivated us to ask: are we applying science seriously by conducting incorrect statistical analyses? The answer is obvious, but we must not be silent, and we must act. We must advocate for conducting the correct statistical analyses in conferences, scientific journals, classroom, online courses, webinars, and formal and informal meetings. We must persuade researchers to use the correct statistical analysis for the benefit of science. The results obtained from a project with wrong statistical analyses might seriously affect nature and humans. Since we understand the crucial role of doing good science, we must advocate using statistical criteria to make the right choice between parametric and non-parametric linear regression. This change is needed to improve our contribution to science and society.

Keywords:

Model assumptions, multiple linear regression, quantile regression, validation.