辅导案例-FTTZ1
FTTZ1 by Zi Wang Submission date: 13-Jan-2020 02:09AM (UTC+0000) Submission ID: 118087904 File name: FTTZ1_1568757_1235762862.docx (134.43K) Word count: 3384 Character count: 19221 Q1.b causal 1 2 3CI 4 R2 Screenreg Q2-e Randomisation 5Changing significance 67 8 Good! Statistical significance Good! Confounding example 9 Statistical significance Causal Confounders 11 12 13 7b treatment effects 14 Missing Intro Univariate stats Nice Plots Plot - labels Explanatory variables 15 16 Missing statistical significance Missing R2 17 18 45% SIMILARITY INDEX 3% INTERNET SOURCES 0% PUBLICATIONS 45% STUDENT PAPERS 1 44% 2 <1% 3 <1% 4 <1% 5 <1% Exclude quotes On Exclude bibliography On Exclude matches Off FTTZ1 ORIGINALITY REPORT PRIMARY SOURCES Submitted to University College London Student Paper Submitted to University of Southampton Student Paper www.economicdynamics.org Internet Source bitsavers.trailing-edge.com Internet Source Submitted to University of Minnesota System Student Paper QM QM FINAL GRADE 44/100 FTTZ1 GRADEMARK REPORT GENERAL COMMENTS Instructor PAGE 1 PAGE 2 Q1.b causal To assess whether this is a causal effect, you should consider if there are confounding factors or differences between treatment and control classes other than their treatment group. Because this experiment involved random assignment, we expect treatment and control classes to be the same, on average, regarding all characteristic, and hence we could take the difference in means as an unbiased estimator of the average treatment effect. Comment 1 this is not correct. check the material for week 7 for the correct formula Comment 2 because of your answer to (c), this t-statistic is also incorrect. however, you show you understand how to interpret it. PAGE 3 Comment 3 same as (d) CI If we took 100 samples of the same size and then calculate their CIs, 95 of them would include the QM QM QM QM QM true average treatment effect. Comment 4 it should be approximately normal, why? PAGE 4 PAGE 5 R2 This question mainly asks about R2, which refers to the percentage of DV's variation been explained by your model PAGE 6 Screenreg You can use screenreg to generate a table for all the models Q2-e For this question, think about the experimental design in this case, and how does a randomised experiment can speak to the potential confounding concerns. Try a t-test of the pre.score for control/treatment group and see if it's significantly different. Randomisation In this case, there is actually no serious confounding concern (although grade and pre.score correlate with post.score) because the randomisation means that these variables are uncorrelated with the treatment by construction thereby ruling out any omitted variable bias in expectation. Therefore, even though the three models shows small differences in the treatment effect, there is not evidence of significant confounding with respect to pre.score as evidenced by the large p-value on the t.test. PAGE 7 Comment 5 missing output table Changing significance how about the change of statistical significance levels for these coefficients? PAGE 8 QM QM QM QM QM QM Comment 6 by how much and is it statistically significant? Comment 7 by how much? Comment 8 it would be better if you make your argument with statistical evidence from the table PAGE 9 Good! Good! Statistical significance Is this significant at 95% confidence level? PAGE 10 Good! Good! Confounding example Democrats are more likely to win state legislative offices in states where they also hold governorship, and Democrat control of these offices may also influence downstream policy outcomes. Comment 9 ...controlling for whether the Democrats control the House and the Senate Statistical significance Is this significant at 95% confidence level? PAGE 11 Strikethrough. Causal We would have to assume that we have controlled for all possible confounding variables to interpret QM QM QM QM We would have to assume that we have controlled for all possible confounding variables to interpret the coefficients causally. Confounders Confounders are likely to be correlated both with the presence of a Democratic govern and with the four policy outcomes. PAGE 12 Comment 11 you need to discuss more the assumptions behind the RDD, the key thing is that the treatment is as good as random. The main weakness is low external validity Comment 12 the numbers are correct but they are the other way around: all the outcomes are lower for democrats than republican Comment 13 you need to plot the two ranges of fitted values separately for dem and rep, the ones you used to find the treatment effects above PAGE 13 7b treatment effects The treatment effects are: -Unemployment: -0.06 -Murder: -0.3 -1% income: -0.12 -House prices: -0.14 Comment 14 They are actually very small PAGE 14 Missing Intro You should have provided a brief introduction explaining the aims and hypotheses of the study. Univariate stats QM QM QM QM QM It would be useful to provide some univariate descriptive stats of all variables included in the analysis. E.g. mean & standard deviation of numerical variables, % of observations in each level of categorical variables. Nice Plots Nice plots Plot - labels use appropriate labels for the X and Y axes Explanatory variables It would be useful to explain why you selected these explanatory variables and how they might influence the relationship between the Catholic % of the population and NSDAP vote share. PAGE 15 Comment 15 You should describe the actual coefficient of Catholic population and also the intercept of the model. Comment 16 You should have compared the coefficient of Catholic population in Model 1 with that found in Model 2. Do the additional explanatory variables confound the relationship between Catholic population and Nazi vote? Missing statistical significance You did not discuss the statistical significance of the regression results. Missing R2 You did not discuss the R2 of the models. PAGE 16 Comment 17 These estimates cannot be causally identified owing to omitted variable bias resulting from the observational design of the study. You should discuss the issue of omitted variable bias in more detail and highlight the limitations of study. Comment 18 You should consider alternative analytical approaches that would help to strengthen causal inference such as RDD and fixed-effects regression. PAGE 17 PAGE 18 RUBRIC: POLITICAL SCIENCE - DEFAULT CRITERION 1 MORE WORK NEEDED (TICK) CRITERION 2 MORE WORK NEEDED (TICK) CRITERION 3 MORE WORK NEEDED (TICK) CRITERION 4 MORE WORK NEEDED (TICK) CRITERION 5 MORE WORK NEEDED (TICK) CRITERION 6 MORE WORK NEEDED (TICK) CRITERION 7 MORE WORK NEEDED (TICK) CRITERION 8 Structure and organisation Development and consistency of argument Extent of research work completed Use of evidence Evidence of critical analysis Overall insight and originality Adequacy and presentation of bibliographic information MORE WORK NEEDED (TICK) CRITERION 9 MORE WORK NEEDED (TICK) Accuracy and appropriateness of referencing Overall presentation