Some of my variables are not normally distributed (with Kolmogorov-Smirnov and Shapiro-Wilk tests). So I used Spearman's correlations. But I also did a linear regression, which shouldn't be the best fit for my data. My question is: if I have significant results using a linear regression, are these findings spurious? Or, knowing that ma data are not linear, if something is significant with a linear regression analysis, it is likely that it would be even more significant with a model that better fits my data? Is this idea correct or totally wrong? And if so, do you have any references to support this?

Thank you

Marie