Hutchinson Paul, Wheeler Jennifer
Department of International Health and Development, School of Public Health and Tropical Medicine, Tulane University, New Orleans, LA 70112, USA.
Stud Fam Plann. 2006 Sep;37(3):169-86. doi: 10.1111/j.1728-4465.2006.00096.x.
Evaluations of national family planning communication campaigns are often complicated by nonrandomized research designs and self-reported measures of exposure to media messages, both of which can bias estimates of campaign effectiveness if suitable analytical methods are not employed. This study uses three methods to address these problems: (1) single-equation multiple regression analysis; (2) propensity score matching (PSM); and (3) maximum-likelihood simultaneous equation models (SEM), namely bivariate probit. We use these methods to evaluate the impact of family planning health-communication campaigns on the use of modern family planning methods, focusing primarily on the radio drama Zinduka! in Tanzania and Ghanti Heri Haad Nilaun in Nepal. We find that these methods led to substantially different estimates of the magnitude of the impact of these communication campaigns. Specifically, we found that self-reported exposure to each radio program was exogenously related to family planning use and that PSM and multiple regression analysis gave similar estimates in those situations. An indicator of exposure to any type of family planning message was found to be endogenous, in which case bivariate probit and PSM gave different estimates of program impact. Researchers should, therefore, pay close attention to the results of tests for endogeneity and identification that guide the appropriate use of the different methods.
对国家计划生育宣传活动的评估常常因非随机研究设计和对媒体信息接触情况的自我报告测量而变得复杂，如果不采用合适的分析方法，这两者都会使宣传活动效果的估计产生偏差。本研究采用三种方法来解决这些问题：（1）单方程多元回归分析；（2）倾向得分匹配法（PSM）；以及（3）最大似然联立方程模型（SEM），即双变量概率单位模型。我们使用这些方法来评估计划生育健康宣传活动对现代计划生育方法使用情况的影响，主要关注坦桑尼亚的广播剧《前进！》以及尼泊尔的《敲响警钟》。我们发现，这些方法得出的这些宣传活动影响程度的估计结果有很大差异。具体而言，我们发现自我报告的对每个广播节目的接触与计划生育的使用存在外在关联，并且在这些情况下倾向得分匹配法和多元回归分析给出了相似的估计结果。发现对任何类型计划生育信息的接触指标是内生的，在这种情况下，双变量概率单位模型和倾向得分匹配法给出了不同的项目影响估计结果。因此，研究人员应密切关注内生性检验和识别结果，这些结果指导着不同方法的恰当使用。
