新的道路基础设施:对企业的影响外文翻译资料

 2022-11-14 15:53:55

New road infrastructure: The effects on firms

Stephen Gibbons /Teemu Lyytikauml;inen /Henry G. Overman /Rosa Sanchis-Guarner

4.3. Robustness checks: additional controls, alternative accessibility indices, spatial autocorrelation and alternative levels of clustering

Table 5 controls for more fixed effects and trends, to check whether results could be picking up growth in business activity that is spuri- ously associated with the accessibility changes, due to the location of the projects –e.g. bypasses on the edges of towns. Column 1 reproduces the baseline results (column 3 of Table 3 ), which control for nearest scheme by year fixed effects and ward fixed effects, plus year dummies interacted with initial accessibility and distance to nearest scheme. Col- umn 2 uses 5 km distance rings from 1–20 km based on distance to near- est scheme, and partials out distance-ring by year fixed effects to control

for time-varying unobservables common to each ring. Column 3 controls more flexibly for nearest scheme by distance ring by year fixed effects, so identification comes from variation in the changes in accessibility oc- curring in a given year, within 5 km width rings around each scheme. Overall, these refinements make little difference. Columns 4–7 augment the column 3 specification with additional ward-level characteristics interacted with year dummies. The first set includes: an urban dummy (more than 90% of the postal addresses are urban); log number of postal addresses, log area, log 1997 employ- ment and plant counts; and log distance to closest Travel to Work Area (TTWA) centroid (as a proxy for the CBD). Column 5 adds an interac- tion between a cubic in distance to nearest scheme and year dummies to control for distance related time trends within the 5 km rings. Finally, columns 6 and 7 introduce 1981 and 1991 census variables interacted with year dummies. These characteristics are all population shares: born in UK, aged 64 plus, over 16 working; owner occupied households; em- ployed in manufacturing, transport, professional and managerial and manual occupations. Again, the idea is to control for time trends in out- comes and accessibility that are related to these census variables. Wealso tried controlling for the corresponding 2001 census characteristics interacted with year dummies, although the 2001 census characteristics are potentially endogenous, coming as they do part way through the study period. None of these modifications have much impact on key re- sults. This reassures us that our design already successfully controls for unobserved trends which might make the accessibility changes corre- lated with business growth, even in the absence of any causal link. Table 6 shows what happens when we use alternative definitions of the accessibility index in Eq. (2) . Columns 1–5 use alternative destina- tion weights ( w k0 ): counts of post office residential delivery addresses (from the ONS NSPD) in column 1; ward population from the 1991 Cen- sus in column 2; ward establishment counts from the BSD in column 3; ward level (residence-based) employment from the 1991 census in col- umn 4; and no weights in column 5, so the accessibility index is simply the sum of the inverse minimum travel time to all potential destinations (within 75 min). The results show that changing the definition of acces- sibility is immaterial, which is not surprising given the high correlations shown in Appendix Table A2 . Columns 6–8 change the proximity function a ( time jkt ) in Eq. (2) from ( time jkt ) minus; 1 . Columns 6 and 7 use alternative parameters in the time- decay function. Column 8 switches to an exponential weighting func- tion. Coefficients are statistically significant for all functions, although they change quite dramatically, reflecting differences in the scale of the accessibility index as the weighting scheme changes: estimates based on one form of accessibility index cannot be used to predict the effects of changes in accessibility based on alternative functional forms. However, standardising the effects (multiply by the standard deviation of the ac- cessibility variables reported in Appendix Table A2 ) gives a more stable pattern of effect sizes. For example, when using inverse time weighting the standardised effect on employment for a one standard deviation in- crease in accessibility is 0.503 times;1.97 = 0.99% (from Table 1 ). For the exponential time decay function in column 8 of Table 6 , the standard- ised effect size is 2.30 times;0.46 = 1.06%. In sum, the results are insensi- tive to changes in the definition of accessibility, or to other specification changes. One remaining concern is that the unobservables may be spatially autocorrelated, leading to biased standard errors and incorrect infer- ence. Given we include ward fixed effects and flexible year fixed effects that vary by closest scheme and distance to scheme, this problem is likely not as important as it seems. We need only be concerned about spatial autocorrelation in the deviations around these fixed effects, not the simple cross-sectional patterns. Direct tests of the regression resid- uals using Moranrsquo;s I statistics take tiny values (less than 0.01) show- ing no evidence of important spatial autocorrelation in the residuals. In addition, re-running our main regressions with errors clustered at a larger geographical level, the Census district, makes very little differ- ence. Table A4 reports results of specification 3 of Table 3 with standard errors clustered in various ways to allow for more general patterns of spatial and temporal correlation. These include clustering at 99 TTWAs,

209 administrative districts, and double clustering at these levels and at scheme-by-year level. The standard errors are similar to those clustered at ward level in Table 3 , column 3.

4.4. Establishment-level employment regressions

The results so far suggest that increased accessibility leads to

剩余内容已隐藏,支付完成后下载完整资料


新的道路基础设施:对企业的影响

Stephen Gibbons /Teemu Lyytikauml;inen /Henry G. Overman /Rosa Sanchis-Guarner

4.3稳健性检查:附加控制,替代可访问性指数,空间自相关和替代的聚类级别

表5控制了更多固定的影响和趋势,以检查由于项目的位置而导致的业务活动的增长是否会促进与可访问性变化有关的业务增长 - 例如。绕过城镇的边缘。第1列再现了基线结果(表3的第3列),它控制了最接近年份固定效应和病房固定效应的方案,以及与初始可达性和距离最近方案的距离相互作用的年度假人。第2列使用距离最近方案距离为1-20公里的5公里距离环,并按距离逐周度控制距离环控制

对于每个环共有的时变不可观测量。第3列通过年度固定效应按距离环更灵活地控制最近方案,因此识别来自给定年份中可访问性变化的变化,每个方案周围5公里宽的环内。总的来说,这些改进几乎没有什么区别。第4-7列增加了第3列规范,其中附加的病房级特征与年假人相互作用。第一组包括:城市假人(超过90%的邮政地址是城市);日志邮寄地址,日志区域,日志1997年就业和工厂数量;并记录距离最近的工作区(TTWA)质心(作为CBD的代理)。第5列添加了距离最近的方案和年度假人的立方距离之间的相互作用,以控制5 km环内距离相关的时间趋势。最后,第6和第7列介绍了1981年和1991年的人口普查变量与年度假人的互动情况。这些特征都是人口份额:出生于英国,年龄64岁以上,超过16岁;业主占用住户;从事制造业,运输业,专业和管理及人工职业。同样,我们的想法是控制与这些人口普查变量相关的结果和可访问性的时间趋势。 Wealso试图控制与年度假人相互作用的相应的2001年人口普查特征,尽管2001年的人口普查特征可能是内生的,因为他们在研究期间已经完成。这些修改都不会对关键结果产生太大影响。这让我们放心,我们的设计已经成功地控制了未观察到的趋势,这可能会使可访问性变化与业务增长相关,即使在没有任何因果关系的情况下也是如此。表6显示了当我们在方程式中使用可访问性索引的替代定义时会发生什么。 (2)。第1-5栏使用替代目的地权重(w k0):第1栏中的邮局住宅交付地址(来自ONS NSPD);第二栏中1991年中心的病房人口;病房建立从BSD第3栏开始计算; 1991年人口普查在第4栏中的病房水平(以居住为基础)就业;第5列没有权重,因此可访问性指数只是到所有潜在目的地的反向最小旅行时间(在75分钟内)的总和。结果表明,改变可访问性的定义并不重要,考虑到附录表A2中显示的高相关性,这并不奇怪。第6-8列改变了方程式中的接近函数a(时间jkt)。 (2)从(时间jkt) - 1。第6列和第7列在时间衰减函数中使用替代参数。第8列切换到指数加权函数。系数在所有函数中具有统计显着性,尽管它们变化非常显着,反映了随着加权方案的变化,可访问性指数的规模差异:基于一种形式的可访问性指数的估计不能用于预测基于的可访问性变化的影响。替代功能形式。但是,标准化效果(乘以附录表A2中报告的可达性变量的标准偏差)可以得到更稳定的效果大小模式。例如,当使用反时间加权时,对于一个标准差的可访问性增加的标准化就业效果是0.503times;1.97 = 0.99%(来自表1)。对于表6第8列中的指数时间衰减函数,标准效应大小为2.30times;0.46 = 1.06%。总之,结果对可访问性定义或其他规范更改的变化不敏感。另一个值得关注的问题是,不可观测量可能在空间上与自相关,导致标准误差和误差不正确。鉴于我们包括病房固定效应和灵活的年固定效应,这些效果因最接近的方案和距离方案而异,这个问题可能并不像看起来那么重要。我们只需要关注围绕这些固定效应的偏差中的空间自相关,而不是简单的横截面模式。使用Morans I统计量对回归残差进行直接测试得到微小值(小于0.01),表明残差中没有重要空间自相关的证据。此外,重新运行我们的主要回归与错误聚集在更大的地理层面,人口普查区,几乎没有差异。表A4报告表3的规范3的结果,其中标准误差以各种方式聚类以允许更一般的空间和时间相关性模式。其中包括99 TTWA的聚类,209个行政区,并在这些级别和逐年级别进行双重聚类。标准错误类似于表3第3栏中病房级别的标准错误。

4.4企业层面的就业回归

迄今为止的结果表明,增加可达性导致病房就业和机构数量增加,至少对某些部门而言如此。就业调查结果可能由现有机构或新机构进入。企业数量结果表明,企业进入可能会促进就业变化,但我们可以通过查看企业内部变化来进一步探讨这一问题。为此,我们使用BSD 1998 - 2008年的企业层面数据估算了原木可获得性对原木就业的影响。表7显示了使用与表3类似结构的关键结果,尽管具有建立而非固定效应。这些回归中的其他控制变量是行业年度假人(上面用于行业结果的6个广泛部门)。和以前一样,错误聚集在病房层面。在大多数情况下,系数是负的,但在所有情况下,它们接近于零并且在统计上无关紧要,这表明交通引起的接入变化不会影响现有机构的就业。与表3一起阅读,其意义在于道路改善通过企业数量和相关就业的净收益促进当地就业,而现有企业保持就业不变。

4.5生产力和其他与生产相关的结果

虽然我们发现现有企业对就业边际没有反应,但如果较低的运输成本允许重组,他们可能会获得生产力提升。我们使用来自ARD的各种输出和输入相关变量直接在表8中进行探讨(在3.2节中描述)。顶部面板显示使用逐年数据的结果,类似于表7。使用来自BSD的逐年特定就业权重对回归进行加权,以使ARD样本更能代表公司群体中的空间雇佣结构。底部面板显示了相同结果变量的病房级聚合的结果,因此可以捕获公司调整时的“密集边际”和建立进入和离开病房的“广泛边际”。在顶部小组中,我们报告了企业层面的劳动生产率的正面影响(每个工人的总产出)。这伴随着工资的增加和对非劳动力投入的需求(这解释了对增值的反应不足)。同样重要的是要记住,1-20公里范围内的平均可达性变化为0.83%,因此这些系数意味着每个工人的诱导产出和公司的工资效应仅为0.20%左右。特定行业的43项结果(未列表)表明,制造业和消费者服务业的影响力最强,这些部门在总体就业方面是一些响应最少的部门,尽管这些影片通常相当混杂且效果不明确。测量。这些影响也集中在拥有10名以上员工的大型企业中。在下图中,我们看到建立,就业和生产率增加的综合影响导致了病房水平生产力的一些相当大的弹性,以应对可及性变化。每个工人的投入成本也有所增加,这与每个工人的产出增加是一致的。这些都是精确测量的。最大和最强的结果是非劳动投入。病房可及性增加1%与非劳动力投入和运输服务总支出增加1.8-2.0%相关。这些病房级非劳动投入弹性的大小(与病房就业和建立水平的投入效应相比)表明,必须对更多生产性企业进行一些分类,将更多非劳动力投入用于改善的地区可行性。我们还采用这些更加综合的结果来表明,随着当地工厂和当地工人数量的增加,表3顶部显示的生产力改进得以持续。结合就业的部门结果,出现了一个图片,其中可访问性改善促使大多数部门的企业进入消费者服务和建筑/能源之外。与此同时,由于中间投入的使用增加,现有企业的人均产出增加。

剩余内容已隐藏,支付完成后下载完整资料


资料编号:[18060],资料为PDF文档或Word文档,PDF文档可免费转换为Word

您需要先支付 30元 才能查看全部内容!立即支付

课题毕业论文、外文翻译、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。