Value-added Based Measures of Teacher Effectiveness and their Determinants in the German National Educational Panel Study

该项目旨在基于德国国家教育小组研究(NEPS)的学生水平成绩数据，分析教师的有效性及其决定因素。现有的美国文献通常使用学生成绩增值模型中的教师固定效应来操作教师有效性，其中估计的教师固定效应作为教师质量的代理。它们量化了教师对学生学习的总体贡献，包括观察到的和没有观察到的教师特质的影响。美国的证据表明，这些教师质量指标的差异可以解释学生成绩总体差异的很大一部分原因。然而，目前还不清楚这种重大影响是否也存在于其他教育体系中；德国目前还没有类似的估计。因此，该项目的第一个关键目标是通过利用NNP数据的纵向结构来估计包括教师固定影响的学生成绩的增值模型来填补这一空白。根据教师质量的估计分布，我们可以量化“有效”教师对学生成绩的影响，通过学生成绩的平均增长与教师质量的一个标准差的改善相关联来衡量。一个特别的目标是将我们的估计与现有的美国证据进行比较，以调查德国教师在有效性方面是否更具同质性。该项目的第二个关键目标是研究教师质量的决定因素。尽管估计高素质教师的影响很大，但界定教师质量的确切特征仍然是一个正在进行的研究课题。非政府组织的调查信息超过了通常在大规模关联的师生数据集中提供的关于教师特征的信息。这使得可以对教师质量的决定因素进行一套新的分析，涵盖三大类教师特征：第一，基本的教师特征，如性别、年龄、经验和教师培训；第二，描述教师态度和行为的变量；第三，关于教师研究过程和教师职业选择的独特非正式数据。例如，我们可以测试接受典型教师教育的教师与决定在以后阶段成为教师的奎因斯坦是否有系统地不同。我们使用两种不同的方法论方法来解决教师质量决定因素的问题，这两种方法都研究了教师特征和学生内部学业成就差异之间的关系，以消除未被观察到的学生和学校特征的影响。第一种方法是基于教师质量的估计增值衡量标准，该方法利用学生成绩随时间的变化。第二种方法利用两门学科之间学生成绩的差异。由于这两种方法是基于不同的识别假设，我们可以基于单一的数据来源来实施它们，我们也将能够评估教师特质的估计影响在多大程度上取决于所选择的估计方法。

The project aims to analyze the effectiveness of teachers and its determinants based on student-level achievement data from the German National Educational Panel Study (NEPS). The existing US literature typically operationalizes teacher effectiveness using teacher fixed effects from value-added models of student achievement, where the estimated teacher fixed effects serve as a proxy for teacher quality. They quantify the overall teacher contribution to student learning, including the impacts of observed as well as unobserved teacher traits. The US evidence suggests that variation in these measures of teacher quality can explain a substantial part of the overall variation in student achievement. It is yet unclear, though, whether such substantial effects are also present in other education systems; no equivalent estimates for Germany exist so far. A first key goal of the project is therefore to fill this gap by exploiting the longitudinal structure of the NEPS data to estimate value-added models of student achievement that include teacher fixed effects. Based on the estimated distribution of teacher quality, we can quantify the effects of having an “effective” teacher on student achievement, captured by the average increase in student achievement associated with a one standard deviation improvement in teacher quality. A particular goal is to compare our estimates with the available US evidence to investigate whether German teachers are more homogenous in terms of effectiveness.A second key goal of the project is to study the determinants of teacher quality. Despite the estimated substantial impacts of high-quality teachers, the precise characteristics defining teacher quality remain a subject of ongoing research. The NEPS survey information exceeds the information on teacher characteristics typically available in large-scale linked student-teacher datasets. This allows a new set of analyses of the determinants of teacher quality, covering three broad categories of teacher characteristics: first, basic teacher traits such as gender, age, experience, and teacher training; second, variables describing the attitudes and behavior of teachers; and third, unique NEPS data on the course of teacher studies and the selection into the teaching profession. For example, we can test whether teachers who followed typical teacher education differ systematically from “Quereinsteiger” who decided to become teacher at later stages.We address the question of determinants of teacher quality using two distinct methodological approaches that both study the relationship between teacher traits and within-student variation in academic achievement to eliminate the effect of unobserved student and school characteristics. The first approach is based on the estimated value-added measures of teacher quality that utilize variation in student achievement over time. The second approach uses variation in student achievement between two subjects. Since the two methodological approaches are based on different identifying assumptions and we can implement them based on a single source of data, we will also be able to assess to what extent estimated effects of teacher traits depend on the chosen estimation method.