Coarse exactly matches the CEM literature recommendation, the program steps can be copied

Everyone who engages in econometrics pays attention to this account.

All the code programs, macro and micro databases and various software of the econometric circle methodology are placed in the community. Welcome to the econometric circle community for exchanges and visits.
Coarse exactly matches the CEM literature recommendation, the program steps can be copied

A few days ago, we introduced IV and Matching veteran, "flexible joint likelihood method" has become a new trend, which introduces related causal inference methods more systematically.
1. Generalized PSM, the best weapon for causal identification of continuous policy variables 2. PSM
tends to match the detailed steps and codes of Stata operation
3. After reading the top journal articles, I compiled the endogenous treatment brochure
4. Treatment effect model selection criteria, NNM And PSM
5. The matching method in causal inference: the most comprehensive review and prospects
6. Endogenity and propensity score matching, a gift for quasi-natural experiments
7. Propensity value matching and causal inference, the most comprehensive and exquisite kit in history
8 .Match or not match? This is really a question worth considering
9. Matching operation guide, 16 articles worth collecting
10. Propsensity matching

In the past few years, we introduced the "CEM matching that does not need to check the balance of covariates, how impressive and unique is it", which has been widely welcomed and discussed by scholars. The article includes detailed operating procedures, steps and data, which scholars can view, download and use by themselves.

Today, we recommend a document using CEM for non-parametric estimation.

The author uses 2013-2014 China Education Survey Group (CEPS) data. Conducted by the National Survey and Research Center of Renmin University of China, CEPS is an ongoing school-based nationally representative longitudinal survey. Using a stratified, multi-stage cluster sampling method, about 20,000 students were randomly selected from 112 schools in 20 provinces of China (10,279 in seventh grade and 9,208 in ninth grade). ceps conducted cognitive tests on students in grades 7 and 9 to measure their cognitive development. This test includes 20 items for seventh grade and 22 items for ninth grade. It comes from the cognitive ability test of junior high school students, and it comes from the non-course analysis ability test of the Taiwan Educational Organization Survey (Yang et al). Year 2003). For students in 7th and 9th grades, cognitive tests are designed to assess their language (for example, linguistic reasoning, phrase analogy), vision (for example, graphical pattern analysis, origami and geometric applications), and mathematics (for example, mathematical words Questions, custom calculation rules, abstract pattern analysis and probability) capabilities.

The main outcome variable that measures the overall cognitive ability of students is generated from cognitive test items using a three-parameter (3pl) IRT model (de Ayala 2013). irt is an important innovation in the field of psychometrics. It is widely used in the design, analysis and scoring of tests, questionnaires and similar tools, which can measure underlying theoretical structures such as abilities, attitudes, and skills. For the author’s purpose, the main advantage of using the three-parameter IRT model to score students’ cognitive abilities is that the score takes into account the different levels of discrimination and difficulty that different test items may have, and that some students may be better at falsehoods than others. The possibility of guessing the correct answer. In addition, the IRT score is independent of the test, which means that even if the author has performed different test items on the 7th and 9th grade students, the author can still combine the two samples for analysis. The IRT score is unscaled, and the approximate normal distribution is zero mean. For more technical details of IRT procedures, see Wang and Li (2015).

The processing variable is whether the respondent is the only child. The author’s analysis also uses school grades (ie 7th and 9th grades), gender, race (Han and ethnic minorities), household registration at birth (urban and rural), immigration history (ie whether the student left the place of birth), and mother’s education , Father’s education, mother’s occupation and father’s occupation. The author also considered the geographic location of the respondents, mainly the province and county where they live.

The author uses the coarsening exact matching CEM (Iacus et al. 2011a, b), which is actually an exact matching plus a preprocessing step, temporarily reducing the dimension to increase the number of matches, as the author's main analysis tool. Unlike propensity scores or other forms of approximate matching, CEM limits the maximum imbalance between the treatment group and the control group through the researcher’s prior selection. This decision can be data-driven, theory-driven, or guided by substantive knowledge. Among the 9 variables used for matching, 5 variables (including school grade, gender, ethnicity, household registration at birth, and immigration) are completely matched, without coarsening, and the education and occupation of the mother and father are coarse before the match is included.化.

The author takes advantage of the flexibility provided by the CEM method and conducts the author's analysis in four steps. In the first step of the analysis, the author's main concern is to assess the average processing effect of the only child (att) of the treatment group on cognitive development. To this end, the author puts children with siblings as a group and paired them with the only child. In the second step of the analysis, the author is interested in testing the potential birth order differences in the cognitive development of only children. In order to do this, the author divides children with siblings into two groups: first-born children with siblings and non-first-born children with siblings, paired with only children. Similarly, in the third step of the analysis, the author is interested in testing the impact of gender differences in the only child on cognitive development. The author followed the advice of Hoet al. (2007) to perform non-parametric matching first, and then use the matched samples for further parametric regression analysis. This method combines the ability of non-parametric matching and the flexibility of regression, providing a simple method to test the effect of heterogeneity treatment. In the last step of the analysis, the author integrates the second and third steps into a set of models, while testing the impact of the birth order and gender differences of the only child on cognitive development. The author conducted a similar analysis, focusing on the order of birth and the hukou status at birth (i.e. rural vs. urban).

In all these analyses, the authors more or less followed the same model building strategy. In each step, the author starts with a model that only matches the 9 variables listed in Table 1. Then in the second model, the author further uses the province id as an additional matching variable (no coarsening). By restricting matching cases from the same province, it is possible to control the characteristics of unmeasured provinces and improve the quality of matching. In the third cem analysis, the author further advanced this idea and restricted the matched cases to the same county (in-county model) by replacing the province id with the county id as an additional matching variable. The use of more refined geographic units in the matching model improves the balance of matching samples at the expense of reduced statistical power. Authors generally prefer results within the county, rather than results from the national sample and within the province, although it is always a good idea to compare different models to determine potential trends.

Comparison of two groups of samples of only children and non-only children.