Using Facebook county-level connectivity data, the US Census Bureau, and the Internal Revenue Services (IRS) information, I find that the traditional determinants of SMP explain the observed cross-county heterogeneity rather well, on average. However, traditional determinants fail to explain SMP heterogeneity across the income distribution. Explanatory power of the empirical model with only the traditional determinants of SMP is particularly low for income-rich households. An empirical model that accounts for the county´s network, namely, when the average SMP rates across the income distribution in the connected counties is included as determinants of SMP, outperforms the traditional framework. SMP rate in the county's network is a particularly important covariate of SMP for the high-income households. This finding highlights the importance of social connectivity in a household´s investment decisions and improves our understanding of nonparticipation among wealthy households.

The model determines how adoption decisions depend on (1) network effects, (2) own learning effects, and (3) social learning effects. We test the theoretical model's findings using unique data from the Bank of Canada's Bitcoin Omnibus Survey for the years 2017 and 2018. After accounting for the endogeneity of beliefs, we find that both network effects and own learning effects have a positive significant impact on Bitcoin adoption, while social learning effects have a negative effect. In particular, a 1-percentage-point increase in the network size increases the probability of adoption by 0.41 to 0.45 percentage points, whereas a 1-percentage-point increase in Bitcoin survival beliefs increases the probability of adoption by 0.43 to 0.55 percentage points. Our results suggest that network effects and individual experimentation were key drivers of Bitcoin adoption in 2017 and 2018.

hshaz2s uses the d2 ml method to provide the algebraic expressions of the first- and second-order derivatives of the log-likelihood function to achieve model convergence faster, which takes special relevance for empirical researchers dealing with large longitudinal microdatasets. Results of fitting a discrete time-duration model that jointly estimates the transition rates from employment and unemployment on a sample of workers in the Spanish labor market are presented to show the main features of the hshaz2s command.

Using longitudinal Canadian provincial-level data over the period 1981 to 2017, we occasionally find a positive correlation between inequality and homicides rates. However, the relationship between income inequality and homicide rates in Canada reverses to become negative when looking at Canadian Census Metropolitan Areas (CMAs). Moreover, the province-level result between greater inequality and homicide rates also appears to break down once accounting for regional effects. We conclude that much of the literature that finds a relationship between greater economic inequality and homicide rates needs to be reexamined within a longer time and more disaggregated framework.

Typically, the method is employed to improve regression estimates from a modest researcher-conducted survey dataset with selected variables in common with a census or large-scale survey. In a generalized method of moments framework, the method simultaneously minimizes the score functions of the estimator of interest (for example, OLS/logit/probit) while also matching suitable moments (for example, means, squares, and cross-products) for the variables common to the survey of interest and the population. As is the case with traditional approaches to generating survey weights, this addresses selection on observed variables that may arise because of issues such as survey design, attrition, and nonresponse. Using Monte Carlo simulation, we show how this method can sometimes improve estimates compared with unweighted OLS and traditional methods of generating weights. We illustrate this method using crowdsource data on COVID-19. The method is easy to use and is also beneficial.

Our goal is to create weights that address three issues. First, the weights should reduce selection bias in each individual survey. Second, the weights should ensure comparability across time of each survey, even if surveys contain different "mixtures" of sample sources. Third, we would like to create separate weights that account for the panel dimension (and therefore attrition) in our April 2020 Survey, where about 1,000 respondents were from past surveys.

While the sequence of models generated this way is unproblematic, the p-values are not valid, because they do not account for the selection procedure. One solution is to compute AIC for the sequence of models generated and to choose the model with the smallest AIC. My command aic_model_selection facilitates this by performing a regression on a sequence of models, adding one x-variable (in the order specified) at a time. I will give an example related to prices of apartment rentals in Munich.

The talk provides a quick introduction to using the lasso and to using the stepwise methods of covariate selection and some tradeoffs between them. It also discusses the authors recommendation to use BIC-based stepwise instead of testing-based stepwise. It also discusses some of the methodology implemented in posw, including some new results in Drukker and Liu (2021).

Using dummy regressions and local projection models, we assess how these indicators of payment habits have changed with the evolution of the COVID-19 pandemic. We find evidence that consumer behavior adjusted to the pandemic by avoiding frequent trips for cash withdrawals and point-of-sale purchases through higher transaction amounts. Consumers withdrew less cash compared with card payments, which could reflect a reduced use of cash for point-of-sale transactions. Consumers also made relatively more withdrawals from ATMs that are linked to their financial institutions (on-us transactions). Finally, we highlight that estimates of economic activity based on card data alone could be biased if shifts in payment habits are not accounted for. We estimate that debit card payments might have overstated consumer expenditure growth by up to 7pp over the course of the pandemic.

Background
As COVID-19 vaccines develop, methods for identifying vulnerability within groups to prioritized vaccination remain unestablished. This presentation describes a novel approach based on population-based analysis of viral pneumonia vulnerability, as an example.

Methods
The analysis employed an anonymous, 16-year population dataset (n = 768,460) consisting of International Classification of Diseases (ICD-9) diagnoses, demographics, and dates identifying those with and without viral pneumonia linked to all associated diagnoses (~90 million) for calculation of independent main-class diagnoses (17) odds ratios and proportions of disorders before and after the index viral pneumonia diagnosis. A subsample of those under the age of 1 year in the first year of the dataset was analyzed prospectively with representation of the intensity of the first 50 diagnoses across all diagnoses (~1000), comparing those who did with those who did not develop viral pneumonia.

Results
Females and males had results of differing magnitude. For those with viral pneumonia, the mean number of diagnoses was greater in both the subsample and whole sample, with associated diagnoses arising about four years on average before the viral pneumonia index diagnosis. Within the subsample, compared with those without, the temporal analysis revealed distinct overrepresentation for those with viral pneumonia at visit 1 and over the first 50 visits. Further, those with viral pneumonia had diagnoses not represented in the group without viral pneumonia.

Conclusions
The population-based analysis of temporal hypermorbidity may be a viable and economical approach to identifying viral pneumonia vulnerability. The approach presented here may provide an economical means of identifying vulnerability to COVID-19 in regions where comparable data are available for analysis. Rational approaches may optimize vaccination and help to limit the spread of the disease and to some extent alleviate the health service burden.

Methods: We queried the 2017 Nationwide Inpatient Sample (NIS) for patients >18 years admitted for a general surgery procedure, comparing outcomes for those with an underlying opioid use disorder with those with none. Opioid use disorder was categorized as abuse, dependence, or poisoning using ICD-10CM coding. The primary outcome was in-patient mortality; secondary outcomes were costs and length of stay. Results: Among 4,445,253 general surgery patients, 157,816 (3.6%) had an opioid use disorder. GS-opioid patients were younger (53 versus 62 years).