Date Published: June 11, 2019
Publisher: Public Library of Science
Author(s): Maryam Zamanian, Farzaneh Zolala, Ali Akbar Haghdoost, Saeide Haji-Maghsoudi, Zeynab Heydari, Mohammad Reza Baneshi, Valerio Capraro.
Network Scale Up (NSU) is a promising tool for size estimation of sensitive issues. In this study we investigated the important methodological considerations to employ this method for estimating behaviors, such as abortion, which happens in a particular age-gender group.
We recruited 1250 males and 1250 females aged 18 to 50. Abortion rate was calculated through direct question and NSU methodology. The NSU was applied on three sub-samples (male, female and aggregate). Integrating replies to 25 reference groups, we estimated the network size (C) of respondents and its age-gender structure. To calculate the part of network that is subject to abortion, we compared two approaches: proportional and data based. The Visibility Factor (VF) was calculated through 222 females who had abortion. Direct estimate was considered as gold standard.
Using C’s derived from proportional method, the Relative Bias (RB) in the male and female samples was 33% and 84%. Applying the data-based C’s, the RB in the gender-specific and aggregate samples was 5% and 2%.
The proportional method overestimates the prevalence. The data-based method to calculate the C is superior. The determination of the age-sex distribution of the network and the specific VF is essential.
Abortion is an important contributing factor in female’s health that could result in irreparable effects and even death of the mother. Abortion is associated with legal restrictions as well as religious and social stigma in many societies [1–3]. In Iran, with the Islamic culture, intentional abortion is banned . In addition, 2012 onwards, the policy of the country has been based on the increase in the fertility rate, which imposed further limitations in practice of abortion. Almost all (98%) unsafe abortions, the third cause of maternal death , occur in developing countries, moreover in contrast to other causes, all complications and deaths related to unsafe abortions could be thoroughly preventable . While the policy-makers need accurate data, to reduce unsafe abortion and to improve maternal health, these barriers make it difficult to obtain valid statistics [7, 8].
Recruiting 1275 male and 1275 female pedestrians, we illuminated how NSU method should be applied in the case of behaviors related to a particular age-gender group. Integrating replies to 25 reference groups, we estimated the network size of respondents and its age-gender structure. It is worth mentioning that there might be overlap between networks of respondents. However, it has been shown that overlapping does not lead to biased estimation .
To our knowledge, this was the first study to perform a multi-way sensitivity analysis to address the influence of the method of the C estimation, the VF and the study population on the size estimation of the age-gender specific hidden characteristics. We have seen that the determination of the age-sex distribution of the network size should be a prime. In addition, the selected sample does not affect the size if the sample-specific parameters are substituted in the formulas.