The confidence interval for Nadaraya-Watson estimator is: $CI_{1-\alpha}{m(x)} = \hat{m_h} \pm z_{1-\frac{a}{2}} \sqrt{\frac{||K||^2_2\sigma^2}{nh\hat{f}_h(x)}} $
I was just wondering if we have a higher value of h does this mean we get thinner confidence intervals? I know when h -> 0 the estimated regression curve will intrapolate between all point and the bias goes to 0 if h --> 0