Title：Practical issues to consider when working with big data（RAS2022）

Abstract：Increasing access to alternative or “big data” sources has given rise to an? explosion in the use of these data in economics-based research. However, in our enthusiasm to use the newest and greatest data, we as researchers may jump to use big data sources before thoroughly considering the costs and benefits of a particular dataset. This article highlights four practical issues that researchers should consider before working with a given source of big data. First, big data may not be conceptually different from traditional data. Second, big data may only be available for a limited sample of individuals, especially when aggregated to the unit of interest. Third, the sheer volume of data coupled with high levels of noise can make big data costly to process while still producing measures with low construct validity. Last, papers using big data may focus on the novelty of the data at the expense of the research question. I urge researchers, in particular PhD students, to carefully consider these issues before investing time and resources into acquiring and using big data.

Summary：In many ways, big data is not inherently different from other types of data. However, researchers, especially PhD students, can forget this in the excitement of learning about a new dataset. This article highlights four practical issues to consider when conducting economics-based research using big data. First, a particular source of big data may not be conceptually different from traditional data. As a result, studies that simply replicate prior results using big data may lack contribution, especially if the new data suffers from the same issues as prior data (e.g., endogeneity). Second, big data sources may only be available for a limited number of entities, especially when aggregated to the unit of interest, leading to limited statistical power and generalizability. Third, high levels of noise and the large volume of data can make big data costly to process; however, an arduous data cleaning process itself does not ensure that empirical proxies are tied to the constructs of interest. Last, interesting research questions are difficult to reverse engineer after the fact, and researchers who invest heavily in big data before generating a research question may end up with a paper that focuses on validating data in the absence of economic intuition. Researchers who keep these four issues in mind can potentially save themselves considerable resources (not to mention heartache!) by avoiding low-impact, high-cost projects and by instead focusing on research questions with the greatest potential for contribution.?

文章結(jié)論：在許多方面，大數(shù)據(jù)與其他類(lèi)型的數(shù)據(jù)沒(méi)有本質(zhì)上的區(qū)別。然而，研究人員，尤其是博士生，在學(xué)習(xí)新數(shù)據(jù)集的興奮中可能會(huì)忘記這一點(diǎn)。本文強(qiáng)調(diào)了利用大數(shù)據(jù)進(jìn)行基于經(jīng)濟(jì)學(xué)的研究時(shí)需要考慮的四個(gè)實(shí)際問(wèn)題。首先，某個(gè)特定的大數(shù)據(jù)來(lái)源可能在概念上與傳統(tǒng)數(shù)據(jù)沒(méi)有區(qū)別。因此，簡(jiǎn)單地利用大數(shù)據(jù)復(fù)制先前的結(jié)果的研究可能缺乏貢獻(xiàn)，特別是如果新數(shù)據(jù)存在與先前數(shù)據(jù)相同的問(wèn)題（如內(nèi)生性）。第二，大數(shù)據(jù)源可能只適用于數(shù)量有限的實(shí)體，特別是當(dāng)匯總到感興趣的單位時(shí)，導(dǎo)致統(tǒng)計(jì)能力和可推廣性有限。第三，高水平的噪音和大量的數(shù)據(jù)會(huì)使大數(shù)據(jù)的處理成本很高；然而，艱巨的數(shù)據(jù)清理過(guò)程本身并不能確保經(jīng)驗(yàn)性的代用指標(biāo)與感興趣的構(gòu)件相聯(lián)系。最后，有趣的研究問(wèn)題很難在事后進(jìn)行逆向工程，研究人員如果在產(chǎn)生研究問(wèn)題之前對(duì)大數(shù)據(jù)進(jìn)行大量投資，最終可能會(huì)得到一篇在缺乏經(jīng)濟(jì)直覺(jué)的情況下專(zhuān)注于驗(yàn)證數(shù)據(jù)的論文。牢記這四個(gè)問(wèn)題的研究人員，可以通過(guò)避免低影響、高成本的項(xiàng)目，轉(zhuǎn)而專(zhuān)注于具有最大貢獻(xiàn)潛力的研究問(wèn)題，來(lái)為自己節(jié)省大量資源（更不用說(shuō)心痛了！）。

通過(guò)www.DeepL.com/Translator（免費(fèi)版）翻譯