Methodology
Methodology
复制标题
DOI：
10.4324/9781003167044-2
复制
发表时间：
2022-02
期刊：
影响因子：
--
通讯作者：
J. Montañez;Amy Donley;Amy Reckdenwald
中科院分区：
文献类型：
--
作者：
J. Montañez;Amy Donley;Amy Reckdenwald
1900-2012年为重叠年份(即这些专家组所编码的年份和专家编码的整个时间段)。更具体地说，我们确定了某一特定国家的全职专家在重叠年份中的经验加权平均分，并从这个值中减去了同一国家的新专家的等价平均分。fia，a，x，a，x，然后，我们在计算先验(限制结果值，使其不能超过顺序数据的范围)时，将此ff差添加到给定国家的新专家的分数中。我们对历史专家使用相同的过程(即，我们分别为新专家和历史专家计算off集)。这些off集的用途如下。对不同时间段进行编码的专家可能在序数尺度的水平上有不同的认知参考点，从而为同一潜在结构提供了由于ff而不同的值。Off集合通过将给定国家-年的先前fiXing设置为一致的参考点，即我们拥有最多数据的专家的分数(对整个时间段进行编码的专家)来改善这一问题。作为进一步的细微差别，从V10开始的新的调查专家对2005年(他们编码任务的fi第一年)之前的一系列年份进行了编码。初步分析表明，这种顺序编码使这些编码者的得分更接近1900-2012年的编码者。因此，在计算经验先验时，我们不会对新的(V10及之后)编码者的分数进行加权。
1900-2012 in overlap years ( i.e. those years both these sets of experts and the full time period experts coded). More specifically, we determine the confidence-weighted average score of the full-time period experts for a specific country in the overlap years, and subtract the equivalent average for new experts of the same country from this value. We then add this difference to the new experts’ scores for a given country for when computing the prior (restricting the resulting values such that they cannot exceed the range of the ordinal data). We use the same procedure for historical experts ( i.e. we compute offsets for new and historical experts separately). The purpose of these offsets is as follows. Experts who code different time periods may have different cognitive reference points for levels of the ordinal scale, and thus provide different values for the same latent construct due to DIF. The offsets ameliorate this problem by fixing the prior for a given country-year to a consistent reference point, i.e. the scores of those experts for whom we have the most data (those experts who coded the full time period). As a further nuance, new experts for surveys starting in v10 coded a sequence of years prior to 2005 (the first year of their coding task). Preliminary analyses indicate that this sequential coding brings these coders’ scores closer to the 1900-2012 coders. We thus do not weight the new (v10 and after) coders’ scores when computing the empirical priors.