I applied poisson regresson on count data. In the model summary I got NULL deviance and residual deviance. I understood that residual deviance is the part which accounts for what is not explained by the model. But I am still not sure of significance of NULL deviance.
The null deviance reflects how well the response variable is predicted based on just an intercept. You can think of this as the null hypothesis, which represents the idea that no model (i.e. intercept only, no covariates) is better than any model you are going to propose. Unless your covariates are completely useless, you should see a smaller residual deviance than null deviance as covariates are added into your model.