However, Levine and Renelt (1992), employing a version of Leamer’s (1983, 1985) extreme bound analysis (EBA), show that growth regression estimates can be very sensitive to small changes in the set of conditioning variables. 1 In order to determine whether findings (1)-(4) from Higgins et al. (2006) are model dependent, we replicate Levine and Renelt’s EBA using the same data set as Higgins et al. We find that 7 out of 11 variables of interest are robust partial correlates with U.S. county-level growth.
Abstract. The paper focuses on the issue of regional resilience against the recent financial and economic crisis in the case of Romania, taking the county as territorial unit of observation. Based on the idea that the shock of a crisis impact spreads asymmetrically in the territory, with different contagion effects, the study advance a new approach of the speed and duration of GDP decline recovering. Data analysis showed that, at macroeconomic level, Romania has not proved resilient to the crisis impact, after two years of recession and a recovery period of 4 years succeeding barely in 2014 to return to the GDP level achieved in 2008. The research highlighted the differentiated recovery duration of the economic decline in territory, in 2014 many counties having to recover in the coming years remained GDP gaps, up to 10 pp or even more. The study paid a specific attention to the crisis impact on employment, focusing on R&D sector as revealing the endogenous growth generating potential at countylevel.
The Jenkins and Chojnacky models were expected to perform similarly, as they are largely based on the same datasets of destructively sampled trees. The two stud- ies combined the meta-analysis differently, partition- ing allometric models from the literature into different combinations based on generalized species classes  or theoretical taxonomic groupings and wood specific gravity . On average, these models produce similar biomass estimates and total county-level predictions, but discrepancies exist on a plot-to-plot basis depending on the species composition of a given plot. Most notably, the Chojnacky models produce greater estimates in high bio- mass plots, potentially because the models used by Cho- jnacky are more species-specific than the Jenkins models.
This county-level study examines factors associated with the rate of voided presidential ballots in the 1996 elections. Evidence indicates that voided ballots are significantly more prevalent in counties with higher percentages of African Americans and Hispanics. The relationship between voided ballots and African Americans disappears, however, in counties using voting equipment that can be programmed to eliminate overvoting. The rate of voided ballots is lower in larger counties, and in counties with a higher percentage of high school graduates. The rate of voided ballots declines as the number of presidential candidates on the ballot increases, but only up to a point, and then rises with further increases. Lever machines generate the lowest rates of voided ballots among types of voting equipment, with punch card systems generating the highest rates.
We combine two forms of target side adapta- tion in this paper: first, we adapt from Facebook to Twitter; next, we compensate for the variation introduced by the fact that Tweets within each county have significant correlation, leading to spu- riously high frequencies of various words in vari- ous counties, significantly reducing the predictive accuracy there of the source models. The remain- der of the paper first motivates our domain adap- tation problem and provides a background on the specific problem of personality prediction. We then situate our method within the field of do- main adaptation. Next, we present the TSDA al- gorithm which, like other popular domain adapta- tion algorithms, is frustratingly easy (Daum´e III, 2007). Finally, we demonstrate that TSDA im- proves the quality of county-level predictions by (a) removing extreme predictions, (b) improving year-to-year stability, (c) increasing average mag- nitude of correlations between predicted county- level personality and measured health and well- being metrics with which the personality con- structs are known to correlate, and decreasing cor- relations where correlations are not expected.
Countylevel nurse densities should also be interpreted cautiously. In very large counties (that are often remote), nurse density, as a metric of access, might be better con- sidered as a product of nurses per unit population and per unit area. The fact that service users may clearly not respect administrative boundaries should also suggest caution in interpreting absolute nurse density values. In- tuitively in geographically very large counties those seek- ing health care are also least able to cross administrative borders in search of services to ameliorate the impact of local low nurse density. Finally it is clear that significant heterogeneity in access to services exists at much more local levels [35-37] and future work should aim to ex- plore access in more detail. Nonetheless the analyses presented are useful at a political and management level as devolution progresses and clearly illustrate the value of developing improved national human resources for health information systems.
Among the collected waste, the greatest share with a clear drop was recorded with respect to non-segregated (mixed) waste with a simul- taneous increase in the level of recovered waste by 4% (Fig. 3). Thanks to the use of recycled materials, pollution of the environment decreas- es, along with the share of primary materials in production, which saves them and does not lead to the degradation of the landscape [Bień and Bień, 2010]. In the mass of segregated munici- pal waste, the dominant group on the level of ap- prox. 23% was glass waste, whose quantity even- tually dropped. A similarly dominant share of such waste was confirmed by Przydatek (2013). Packaging waste, according to Rosik-Dulewska (2015), constitutes one of the major problems of municipal waste management. An increase by 6.61% in the share of municipal waste col- lected selectively referred to plastic packaging. A similar increase in domestic conditions was shown by Szymańska-Pulikowska (2012). On the other hand, other researchers determine the changed consumption model and resignation from the system of return packaging subject to
The production and release of chemicals into the air and waterways can result in negative health outcomes for exposed populations. These releases come from vehicle exhaust, power generation, and the numerous companies that use or produce chemicals. In order to capture multiple aspects of the pollution in a county, two measures of air quality along with the total volume of releases are utilized in this research. Air quality measures are the number of days per year with unhealthy levels of ozone and the number of days per year where particulate matter is at an unhealthy level. Both of these measures have shown links to numerous health concerns, including cancer and are part of the clean air index (EPA 2009). The Toxic Release Inventory (TRI), maintained by the United States Environmental Protection Agency (EPA), provides chemical release information and includes all substances stored or released that pose harm to human health. The EPA, due to their known toxicity, regulates and tracks these substances. The EPA data is composed of countylevel counts of TRI emissions as raw counts as well as the count of days with unhealthy air. (Table 3.1)
Alcohol-impaired driving has been a serious public health problem for decades [4, 7]. Despite previous estimates of over 100 million annual episodes of alcohol- impaired driving nationwide, only approximately 1.1 million arrests were made for driving under the influence (DUI) in 2014 [8, 9]. Although previous reports on alcohol-impaired driving provide national- and state-based estimates, considerable injury prevention and traffic enforcement occur at the municipal level; thus, county-level information is useful to help guide allocation of limited public health resources and to support targeted injury prevention efforts. Using data from the Behavioral Risk Factor Surveillance System (BRFSS), we sought to estimate the county-level annual prevalence of alcohol-impaired driving in every US county for the years 2002 through 2012.
As this study shows, ring maps can highlight racial disparities in health, convey epidemiological uncertainty data (e.g., confidence interval data associated with stan- dardized morbidity and mortality rates), and suggest small area-level associations between adverse health out- comes and characteristics of the socioeconomic and built environment. The ring maps presented here only begin to illustrate the potential utility of this visualiza- tion method for health geographers. For example, ring maps can depict multiple attributes at the census tract, census block group, ZIP code area, hospital catchment area, or public health service area level, in addition to the countylevel as shown in the figures [25,27]. In addi- tion, a ring map can be used to depict a single attribute at multiple geographic scales. For instance, a ring map might show diabetes prevalence rates for South Carolina at the census tract level in a base map and prevalence rates at the countylevel and public health region (multi- ple county) level in successive rings (in this case, the number of enumeration units in the inner ring would reflect the number of counties, and the number of enu- merations units in the outer ring would reflect the num- ber of health regions in the state). Ring maps also can display time-series data for a single variable of interest [25,27]. A map with six rings might show annual inci- dence rates of cardiovascular disease over a six-year per- iod, for example; alternatively, a ring map might depict
At countylevel, in the year 2013, the most important weights in Romania’s exports (Table 3) were held by the Bucharest municipality (17.36%), Arges county (10.25%), Timis (9.38%), Arad (5.07%), Constanta (4.92%), Brasov (4.64%), Sibiu (4.08%), Prahova (3.38%). These counties maintained for the analysed period their ranking in the national top of exporters, with slight differences, and were noticeable for putting to good use the local endogenous potential (existing endowment, local labour force, localisation advantages, etc.) that attracted the interest of foreign investors, who are present to a large extent in the respective regions.
The conjoined framework is applied to the U.S. labor market from 1998-2015. Namely, we inspect the dynamic roles of diversity/specialization and modularity across both geographic and industrial scale. We find regions with a roughly equal representation of all major industries complemented by county-level specialized industry clusters is, on average, the most resistant and recoverable structural form that cultivates low unemployment rates across economic cycles. As a result, the suggested optimal labor market structure is industrially diverse and self-similar at higher scales (nation/region) and specialized at lower scales (state/county). As an aside, during periods of economic expansion, industrial migration between major industries at the state level is seen to be correlated with lower unemployment rates. This result reverses during periods of recession, which suggests an evolutionary dynamic within the labor supply. Overall local levels of unemployment rates are strongly tied to the features of a robust labor market. Likewise, these features seem to comfortably fit within the regional resilience framework as possible determinants of system dynamics.
If an omitted variable causes correlation between obser- vations across counties, then the error terms would also demonstrate this correlation. I address this potential issue by clustering standard errors at county-level in all my regressions. However, there are large differences in the underlying population in each county and this clustering approach may not be valid . One way to check robust- ness of my results is to cluster standard errors by state, although clustering on only a few states is not always opti- mal . I report the results with standard errors clustered at state-level in the last column of Table 7 where all the coefficients still have the same sign and significance level. Therefore, I rule out the possibility that the precision of standard errors is driving the results.
We conducted individual interviews with 269 individuals and 14 focus group discussions with a further 146 par- ticipants (see Table 4 ) between March 2015 – April 2016. Participants include 14 purposively selected na- tional level key informants with specialist knowledge of the health priority-setting process. We purposively se- lected 120 countylevel decision-makers from ten diverse study counties (see Tables 3 and 4 ) to include a range of actors involved with priority-setting, including: politicians involved with decision-making for health, county treasury staff, gender and children’s office repre- sentatives and technical decision makers for health in- cluding members of the county health management team. We continued to interview across counties, due to the diversity of contexts and continued until saturation was reached by respondents at countylevel. Saturation was considered reached when there were no new major themes emerging from the findings. In-depth interviews (IDIs) with 49 health workers from sub-county, health facility and community levels were carried out in three (out of the ten) counties. Due to research time and re- source constraints it was not possible to conduct this depth of research across all ten counties studied for the countylevel interviews, however, we sought to ensure as much diversity of responses as possible, by including counties which represented urban, rural agrarian and rural pastoralist settings.,We carried out interviews with 86 close-to-community (CTC) providers, their supervisors and community members and 14 focus group discussions with community members from two counties (out of the three) (see Table 4 ). This data was collected as part of an ongoing REACHOUT CTC provider quality improvement study in two counties (urban and rural agrarian). REACHOUT is an ambitious five year international research consortium aiming to generate knowledge to strengthen the performance of CHWs and other close-to-community (CTC) providers in promotional, preventive and curative primary health services in six
The administrative authority that is closely related to mergers and reorganizations, and is a unit according to the administrative area, is the authority to formulate the mergers and reorganizations plan. The main authority for carrying out the campaign was decen- tralized to local governments on the basis of localized management theory. This in- cluded three types of administrative authority: 1. The authority to determine enterprises qualified to be main merging and reorganizing parties; 2. The authority to decide the coal mines to be merged (or closed); 3. The authority to approve limits on merging for main merging enterprises. Three methods were then used to allocate these three types of administrative authority within each province: the first method is that these three types of administrative authority were mainly held by provincial govern- ments, which in principle, were not given to prefectural (county) governments. We de- fine this situation as “ decentralization to provincial level ” . 12 The second is that the provincial government only identifies the target and the basic principle of the cam- paign, while the three types of administrative authority are allocated to the prefectural (county) government. In this situation, the prefectural (county) government is respon- sible for planning the mergers and reorganizations implementation schedule, as well as organizing how to carry it out. This is known as decentralization to prefectural (county) level. The third method lies between the previous two, where a considerable part of ad- ministrative authority is allocated to the prefectural (county) government, while the rest Table 2 Basic features of the mergers and reorganizations policy for each province (including autonomous regions and municipalities) (Continued)
2016 . Countylevel human CDC cases were retrieved from the CDC through the National Notifiable Disease Surveillance System . Countylevel population data was included from the US Census Bureau decennial census in 2010 . The negative binomial model was chosen due to the nature of the data (count data), distri- bution of the data (right-skewed distribution), and over dispersion of a Poisson model (p < 0.0001). All Northeast- ern counties were used in the model, including counties with no I. scapularis submissions. Model fit was assessed through Akaike Information Criterion (AIC) , mean absolute error (MAE), Root Mean Square Error (RMSE), and Normalized Root Mean Square Error (NRMSE) was normalized based on the mean of CDC reported Lyme disease cases. Additionally, model fit was assessed based on the location of each county by US Census Division (New England and Middle Atlantic) and by state. These geopolitical lines do not have an impact the ecology of the system, however, variations occur in state reporting guidelines for tick-borne diseases. For example, Massa- chusetts has changed the reporting guidelines for Lyme disease, resulting in a 20× decrease of reported Lyme disease cases between 2015 and 2016 (2015: 4224 cases, 2016: 198 cases) [5, 49].
possibility that increased custody rates reflected an institutional response to increased delinquent activity. In addition to the overall juvenile arrest rate, the juvenile violent crime rate was included in some analyses because secure custody is most often used in cases involving serious crimes. County-level juvenile arrest data specifically for violent crime were not available for the entire time period. As a proxy, the overall violent crime rate was discounted by the proportion of all criminal arrests involving youth ages 10-17 in each county for a given year. Because juveniles tend to commit violent crime at a lower rate than other crimes, the measure probably overestimated the level of juvenile violent crime. If the use of secure custody is a response to fear of crime, rather than actual crime by juveniles, it seems plausible that the overall violent crime rate would also affect custody rates because most violent crime is committed by adults. The measure utilized represents a level of violent crime that likely falls between the actual levels of juvenile and overall violent crime. The greater prevalence of adult crime as compared to juvenile crime may influence perceptions about criminal behavior and influence juvenile punishment practices, so the overall crime rate was also included in some models. Each of these three crime measures had observations with missing data, so the total analytical sample included 1,557 observations.
This table reports the results of the analysis of the effect of minimum wage on corporate innovation. The sample consists of firms located within 50 km of the borders of contiguous county-pairs. County-pairs that straddle two provinces are excluded. Corporate innovation is measured using a firm’s patent information. The dependent variable, ln(P atent), is defined as the log value of one plus the number of new patents applied for by the firm in the focal year. ln(M W ) measures the (log) annualized minimum wage based on the monthly minimum wage in December of the focal year. Column 1 presents the baseline regression using the baseline sample. We repeat in Columns 2-7 the sample split approach described in Tables 5-7 to analyze the heterogeneous effects of MW on innovation according to labor intensity, access to credit, and industry competition. We include as control variables the log value of lagged total assets (ln(Assets)); lagged tangibility measure (T angibility); profitability measured by return on assets (ROA); and growth opportunity measured by sales growth ratio (∆Sales). The city-level economic variables include the growth rate of GDP (∆GDP ), level of GDP per capita (ln(GDP per capita)), and foreign direct investment growth rate (∆F DI). Also reported are the p-values for the null hypothesis that the coefficient estimates of ln(M W ) are the same in Columns 2 and 3, Columns 4 and 5, and Columns 6 and 7. All of the regression specifications control for fixed effects at the province-year, industry-year, county pair, and firm level. t-statistics are based on robust standard errors clustered at the county-pair level and reported in brackets. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.
tricts and counties, between 8.26 ∼ 51.52 %. Significant in- crease in a larger proportion for Zhenba County, Ziyang County, Xunyang County, Zhashui County, Shangzhou Dis- trict, Shanyang county and Yun county, were more than 20 %. The largest is in Shangzhou District, and it reached 31.11 %. Overall, the NDVI increased proportion in each county is far greater than the proportion of reduced. The reason for the decrease of vegetation index in Hantai is that in recent years, Hantai district has been actively developing and building in- dustrial parks, and the total area of cultivated land has de- creased by 16.56 % from 2000 to 2016.