sjtu-cxy

Author: cxy0714

notes/Master.bib

@@ -1,3 +1,12 @@
+@article{levis2025covariate,
+  title={Covariate-assisted bounds on causal effects with instrumental variables},
+  author={Levis, Alexander W and Bonvini, Matteo and Zeng, Zhenghao and Keele, Luke and Kennedy, Edward H},
+  journal={Journal of the Royal Statistical Society Series B: Statistical Methodology},
+  pages={qkaf028},
+  year={2025},
+  publisher={Oxford University Press UK}
+}
+
 @article{wu2025promises,
   title   = {The Promises of Multiple Experiments: Identifying Joint Distribution of Potential Outcomes},
   author  = {Wu, Peng and Mao, Xiaojie},

notes/main.typ

@@ -168,6 +168,26 @@ The potential outcome model is an example of latent structure model. The observe
 
 === Bell inequality
 
+
+== Instrumental variable
+
+@levis2025covariate 
+- data are $(X,A,Z,Y)$ only assume consistency, positivity, unconfounder, exculusion, no monotonicity, provide a bound estimation on ATE.
+
+- The identification assumption of IV:
+#quote([ Critically, under the four assumptions introduced in the previous section, the ATE is not point
+identified. Analysts typically take one of two approaches for point identification. The first
+approach invokes some type of homogeneity assumptions and places various restrictions on
+how the effects of A and Z vary from unit to unit in the study population. See Hernan and
+Robins (2019) and Wang and Tchetgen Tchetgen (2018) for prominent examples. However,
+homogeneity assumptions are often implausible or difficult to verify in specific applications.
+The second approach invokes an assumption known as monotonicity, which has the following
+form: A(z = 1) ≥ A(z = 0), i.e., if A(z = 0) = 1 then A(z = 1) = 1 (Imbens and Angrist,
+1994). Under monotonicity, the target estimand is no longer the ATE, but instead is the local
+average treatment effect (LATE):])
+
+- The lower bound and upper bound is not a differentiable functional, thus an assumption is invoked to make the bound functional differentiable and thus have inference function to faster convergence rate.
+
 = Semiparametric theory
 
 == Parametric theory

update-notes.sh

@@ -12,8 +12,10 @@ TYP_FILE="notes/main.typ"
 OUT_DIR="static/notes"
 OUT_FILE="${OUT_DIR}/notes.pdf"
 
-# === Windows 下 Typst 可执行文件 ===
-TYPST_WIN_PATH="/c/Users/thinkbook-cxy/AppData/Local/Microsoft/WinGet/Packages/Typst.Typst_Microsoft.Winget.Source_8wekyb3d8bbwe/typst-x86_64-pc-windows-msvc/typst.exe"
+
+# === Windows 下 Typst 可执行文件（更稳健版本）===
+TYPST_WIN_PATH="$(cygpath "${APPDATA}/../Local/Microsoft/WinGet/Packages/Typst.Typst_Microsoft.Winget.Source_8wekyb3d8bbwe/typst-x86_64-pc-windows-msvc/typst.exe")"
+
 
 # === 检查 Typst ===
 if [ -f "$TYPST_WIN_PATH" ]; then