# R: linear regression: very small coefficient and R-squared but significant P values

I've got a very small coefficient (-0.04) and R-squared (0.028) but a significant P value (<0.0001). My question is:

• Is my result still meaningful?
• How to interpret it?

The result is from a linear regression model in a big database in R. The independent variable (B) has more than 200 values, whereas the dependent variable (A) has 13 values.

The potential correlation is below: So, I ran a linear regression model between A and B and the result is:

>mod1<-lm(A~B)

Call:
lm(formula = A ~ B)

Residuals:
Min      1Q  Median      3Q     Max
-63.174 -11.816  -1.651  10.184 118.001

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 132.274547   0.303723 435.511  < 2e-16 ***
B            -0.036675   0.009052  -4.052 5.13e-05 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 18.52 on 8093 degrees of freedom
(123 observations deleted due to missingness)
Multiple R-squared:  0.002024,  Adjusted R-squared:  0.001901
F-statistic: 16.42 on 1 and 8093 DF,  p-value: 5.134e-05


As you can see, the coefficient of B is only -0.03 and adjusted R squared is only 0.1% but with a p value <0.0001. Is my result reasonable and countable? Surely, my database is larger (8000 records) and even a very small effect size will show a significant P value. But how would I interpret this?

• "whereas the dependent variable (A) has 13 values" Does that mean A is not a continuous variable? Is B a continuous variable? How do you define "reasonable and countable"? – Roland Nov 9 '16 at 15:20
• In your plot it looks like B has 13 values and A has more than 200 different values instead of the other way around. – Pieter Nov 9 '16 at 15:32
• It is very hard to think of any real situation, where this small an effect is going to play any role in prediction. The effect is tiny, so tiny, that you can not spot it with the eye, yet with a large n you can proove, that the slightest trend exists. Is it meaningful? Does it tell the science community anything? That can not be answered without knowing what you examined an know more about your topic of study. Likely? No! – Bernhard Nov 9 '16 at 15:32
• The R^2 and p-values depend on assumptions of normality which are violated here, so I wouldn't place any stock in the result. – Paul Nov 9 '16 at 15:43
• @Paul those depend on assumptions of normality of the residuals. – Firebug Nov 9 '16 at 15:58