In the research design, MLRA was proposed since relationship is to be determined between several independent variables and a dependent variable. Nevertheless, the results was found to be not significant in all independent variables even at .05 level. It was pointed out the MLRA is not the right statistics, thus it was recommended that the statistical tool be changed because the non-significance means that linearity was not satisfied in the first place.
Ernie Celino Cerado, I think there should not be any problem with changing your method of analysis, particularly because, under the circumstances, it seems you'd chosen an inappropriate method of analysis originally.
You do not seem to be hunting around with different methods until you achieve statistical significance - which would be a very undesirable "fishing" practice.
It is not mandatory that we should always get a significant relationship always. However, changing the suitable statistical method justified provided it is carried out with honesty and good motive (without manipulation).
Thank you for the insights that you have provided as regards my query. I asked this because it has become a notion (due to repeated comments and suggestions) that whenever MLRA fails or do not obtain significant results, it also means that the statistical design is wrongly thought of.
Ernie Celino Cerado, I am sure that multiple linear regression often produces nonsignificant results because there is no relationship between the variables being analysed.
That aside, there are certain assumptions that need to be met for MLRA (I'll provide a site regarding this below), and, if those assumptions are not met, either Type I or Type II errors might occur.
Here is a site about the assumptions for MLRA:
https://www.statisticssolutions.com/assumptions-of-multiple-linear-regression/?__cf_chl_jschl_tk__=bc54ebbe2e08808f9c79db2ef5ba950de033cadc-1616410177-0-AfZWiccsaRtWgc94ChzogJkqEpJG53e1fsvdLoXCfpJjGkm4SipbNrktrjyG4AhTYnOVx0t8ScfLMZKnxlLyXAVROMKH966OrWWi1yIN5pYlGFfNOSICRa99elPlg15tgT9xkDWL6n6J60yKHkvba3ulU54MZjk5AKjYu00YchI2HUDIbdcodx9qZ59F_7BVGUosh2j746VqtAZZkPFe4xnXChvxJMAPKPSHNMkPP716MM-FFlgrNkdnYOsuPo0zP99fKRYrv9qeCBRTt06zGLRggoV1t44ALkCajIxEUuCBz-cr_jAYVQ89Ff5AkZ9GOSYy_hL5tkHlFPdAR87mGZAtWi_c7PhetSn2r-MlL0gVZG3VOOQikTW3dxXuWT9kRg
You state that the reason for the failure of the regression technique was because "non-significance means that linearity was not satisfied in the first place." But there are many ways to use regression to test whether the independent variable had a non-linear effect. The idea that the effect of a variable is linear is simply an initial hypothesis, and you can you extend your analysis to test a variety of alternative hypotheses about non-linear effects (one common example is to add an X-squared term to test for a quadratic effect).
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers