A note on composite likelihood inference and model selection