勞動市場集中度對薪資的影響

Abstract

2018年北漂成為國內熱門的選舉議題, 不同區域的就業機會差異是導致勞動力移動的重要因素之一, 但國內對於薪資區域差異的影響因素探討甚少。考慮不同區域間雇主買方壟斷差異導致的就業機會差異,本研究探討臺灣各勞動市場之集中度,並運用多元線性迴歸模型分析勞動市場集中度對於薪資之影響。本研究爬取104人力銀行2018年4月至2019年6月職缺資料,並以通勤區 — 職業別 — 每季定義勞動市場範圍,運用赫芬達爾 — 赫希曼指數 ( Herfindahl-Hirschman Index, HHI )作為計算勞動市場集中度的指標。研究結果發現, 臺灣各勞動市場 HHI 平均達3,189,依據美國法務部及聯邦貿易委員會指引,屬於高度集中的勞動市場,且都會區 (六都) HHI 平均為2,101 (中度集中),非都會區 (六都以外縣市) HHI 平均為3,970。迴歸結果發現,當控制職業別、職缺所在產業類型、職缺教育程度需求、職缺年資要求、失業率、區域生活成本、時間及區域固定效果後,在非都會區及相對低薪（即薪資未達4萬）之職缺,HHI 對薪資有負向且顯著之影響,此實證結果亦具穩健性。

In the 2018 Taiwan municipal elections, “north drifting” became a hot topic. The difference in employment opportunities among different regions is one of the main factors causing the labor force mobility; however, there are few and limited related researches in Taiwan. This study explores the labor market concentration in Taiwan in connection with the difference in employment opportunities caused by the different monospony degree of employers among different regions. This study calculated labor market concentration using the Herfindahl - Hirschman Index (HHI) for each commuting zone by occupation and season based on job data from April 2018 to June 2019 crawled from the 104 online job bank, the largest online job board in Taiwan. The average market is highly concentrated (above 2,500 HHI) with an HHI of 3,189, the 6 municipalities have an average HHI of 2,101 (moderately concentrated), and counties or cities other than 6 municipalities (non-urban areas) has an average HHI of 3,970 (highly concentrated), according to the United States DOJ/FTC guidelines. Furthermore, the regression using multiple regression analysis methods for jobs that are relatively low-paid ( salary less than 40,000 NTD ) or located in non-urban areas showed that HHI has a significantly negative impact on wage when other possible impact factors of occupations, industrial types, job education level requirements, years of work experiences, unemployment rates, regional living costs, time and area fixed effects were controlled. Also, a number of additional robustness checks were performed, and the results indicate that the findings are robust.