I have used the synthetic_us_2010 dataset to reproduce the results, and now I have three questions

[QAZqaz123wkx]

Hello, I have used the functions in the CausalGPS R package and the synthetic_us_2010 dataset to reproduce the results, and now I have three questions for you:
First，I cannot find an attachment in the R package that can check whether my reproduction results are correct or not, may I know where I can go to check my reproduction results?
Second, The article 'Matching on Generalized Propensity Scores with Continuous Exposures' mentions that you can use a data-driven (grid search) approach to select the parameters that achieve the best covariate balance (δ, λ), I don't know how to realize this, could you please point me in the right direction?
Third, In the part of result visualization, I only reproduce the smoothed dose response function, how can I reproduce the smoothed causal effect plot?

[wxwx1993]

Hi @QAZqaz123wkx, Thank you for your interest.

1. We recommended using simulated data to check the usability of the methods (https://nsaph-software.github.io/CausalGPS/articles/Generating-Synthetic-Data.html). The synthetic_us_2010 is limited in sample size, only providing an illustration as a case study. Alternatively, you could check the application section of https://arxiv.org/pdf/2310.00561. The data set is at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/5XBJBM

2. You may use grid search on (δ, λ) via a loop function outside of generate_pseudo_pop. I recommend that you fix λ = 1 and choose δ for reduced computing needs. Also see https://arxiv.org/pdf/2310.00561.

3. Could you clarify what you mean by the smoothed causal effect plot? Since the dose-response curve is Y(a) throughout a range of a. If you refer to causal effects as Y(a1) - Y(a2) and/or Y(a1)/Y(a2), then it can be derived from the dose-response curves.