Seminars

A data-driven approach to cell ratio imputation for item nonresponse in data linkage problem

Wednesday, February 6, 2019 - 12:10pm to 1:00pm
Event Type: 

Danhyang Lee / Abstract: With the improved availability of administrative data sources, data linkage, which integrates administrative records with survey data, has been increasingly used to improve the quality of official statistics. However, there are some limitations in directly using the linked data for statistical analyses when administrative data does not cover a whole population of interest, hence, the linkage between survey data and administrative data is not perfect. Read more about A data-driven approach to cell ratio imputation for item nonresponse in data linkage problem

Progress Report: Visualization of Sheet and Rill Erosion on US Cropland

Wednesday, February 13, 2019 - 12:10pm to 1:00pm
Event Type: 

Xiaodan Lyu / Abstract: National Resource Inventory (NRI) is a longitudinal survey which monitors national resources on non-federal US land. It provides annual national and state-level estimates of average water erosion on cropland. Natural Resources Conservation Service (NRCS) did some primitive attempt to visualize sheet and rill erosion rates on US cropland every 5 years. Their maps represent a surface model based on a neighbor interpolation of all NRI sample sites. Read more about Progress Report: Visualization of Sheet and Rill Erosion on US Cropland

Summary of 2017 MMDS Survey

Wednesday, February 20, 2019 - 12:10pm to 1:00pm
Event Type: 

Zhenzhong Wang / Abstract:  Pet Demographics Survey (PDS) is a nationwide survey on pet ownership, veterinary visit and expenditure, which is conducted every five years. Metro Market Demand Survey (MMDS) is a smaller survey conducted every year. It is a supplemental survey for PDS and surveys about six geographic sub-regions of the U.S.. It has two goals: (1) Provide estimation of household pet ownership, veterinary visit and expenditure of each sub-region. (2) Annually update the nationwide PDS estimates by imputation. Read more about Summary of 2017 MMDS Survey

Variance Estimation for Imputed Data

Wednesday, February 27, 2019 - 12:10pm to 1:00pm
Event Type: 

Xiaofei Zhang / Abstract: Imputation can be used to complete records for a subsample of a large sample, when the subsample of the large sample has incomplete records. Data may be missing due to nonresponse or due to design. We give several mean estimation procedures based on nearest neighbor imputation for both nonresponse and design missing. We also give design consistent variance estimators for the mean estimators. Read more about Variance Estimation for Imputed Data

Functional Covariate Adjustment for Calibration

Wednesday, March 6, 2019 - 12:10pm to 1:00pm
Event Type: 

Hengfang Wang / Abstract:  Calibration estimation is widely used when researchers want to estimate the population mean with the adjustment of the survey weights when auxiliary variables are available. However, before such calibration estimation, several functions for calibration equations have to be determined. In this talk, we employ a reproducing kernel Hilbert space to do the approximation between sample space and population space of covariate to get such adjustment of survey weights without predetermining functions. Read more about Functional Covariate Adjustment for Calibration

Subgroup Analysis over Networks

Wednesday, March 13, 2019 - 12:10pm to 1:00pm
Event Type: 

Xin Zhang / Abstract : We consider an adaptive fused-lasso based coefficient subgroup approach in the decentralized network system. The major goal is to improve the model estimation efficiency by aggregating the neighbors' information as well as identify the subgroup membership for each node in the network. In particular, a tree-based subgroup penalty is proposed to save the computation and communication cost. Also, we design a decentralized generalized alternating direction method of multiplier algorithm for solving the objective function in parallel. Read more about Subgroup Analysis over Networks

Satellite data based surface water classification and its application to NRI data collection

Wednesday, March 27, 2019 - 12:10pm to 1:00pm
Event Type: 

Charlie Labuzzetta / Abstract : Land cover is one data type reported for sample points in the National Resources Inventory (NRI). We present a satellite data based classification pipeline for assessing the presence of surface water land covers at NRI sample points. This pipeline includes data acquisition using Google Earth Engine, STFit imputation of missing values, Random Forest based classification, and outlier detection. Various methods of visualization of such classifications and analyses will also be presented. Read more about Satellite data based surface water classification and its application to NRI data collection

Remote sensing feature detection using deep learning model - BMP project case study

Wednesday, April 17, 2019 - 12:10pm to 1:00pm
Event Type: 

Lei Zhou / Abstract : Iowa Best Management Practices (BMP) Mapping Project was originally aimed to monitor the nutrients reduction measures in Iowa in accordance with the 2008 Gulf Hypoxia Action Plan. In doing so, it generated six conservation practices labels (e.g pond dam) through human labeling by using remote sensing images. Our project tries to generate pond dam labels as an application demonstration by leveraging deep learning models and massive training images with labels. Our model shows that the labeling step could potentially be automated. Read more about Remote sensing feature detection using deep learning model - BMP project case study

Bivariate Small Area Estimation with Monte Carlo EM algorithm

Wednesday, April 24, 2019 - 12:10pm to 1:00pm
Event Type: 

Hao Sun / Abstract : Small area estimation (SAE) is widely used to produce reliable estimates of characteristics of interest such as means,counts, quantiles when there exists some limitations of the available data. In SAE, people usually focus on univarite response with generalized linear mixed model to which allows for additional variability. However, in many cases, we could have bivariate correlated responses and fit GLMM for them together will be very tough. Read more about Bivariate Small Area Estimation with Monte Carlo EM algorithm