| Category | ORIGINAL_ARTICLE |
|---|---|
| Authors | Adhin Bhaskar, Kandavel Thennarasu, Mariamma Philip, Supraja Thirumalai Ananthanpillai, Geetha Desai, Prabha Satish Chandra |
| Abstract | Background: Multimodality, the occurrence of multiple modes in the distribution of the data is a less-discussed issue in the area of count regression models. Objectives: To assess the suitability of Hermite regression model to predict the factors affecting the number of symptoms among pregnant women. Methods: This study is focused on comparing the performance of various count regression models when the dependent variable is both overdispersed and multimodal. The data is obtained from a community based prospective study of anxiety, depression during pregnancy and its relationship to pregnancy outcomes. Poisson, negative binomial, Hermite and generalized Hermite regression models were fitted to find the relationship between the variables. The models were compared using fit indices along with the estimates and standard errors. Distribution of randomized quantile residuals was also assessed to determine the goodness of fit of the models. Results: Based on the values of fit indices and tests, Hermite regression was chosen as the best to establish the relationship between the response variable, number of somatic symptoms and the predictors. The model identified parity, stress and depression as the factors affecting number of somatic symptoms in pregnant women. Conclusions: The Poisson and negative binomial model may not accommodate multimodality as they are framed based on unimodal distributions. The Hermite regression approach is an ideal approach for count data, as it can handle both overdispersion as well as multimodality. |
| Year | 2020 |
| Month | July |
| Volume | 9 |
| Issue | 3 |
| Published On | 31 Jul 2020 |
| Read PDF | |
