Hi Eddie,

I think the secondary analysis is recognized by research community as a scholarly research method / strategy.  Because I remembered saw some articles being published in some journals.  Off hand I can't retrieve those articles.  But the research areas surrounding stock market e.g. prediction of Asia financial currency crisis by using some historical economics data (secondary data), evaluation of company specific performance by using secondary data like their shareholder annual report etc.  I also recalled those secondary data adopted need to be justified for its validity, reliability & recency.

Hi Eddie,

I think the secondary analysis is recognized by research community as a scholarly research method / strategy.  Because I remembered saw some articles being published in some journals.  Off hand I can't retrieve those articles.  But the research areas surrounding stock market e.g. prediction of Asia financial currency crisis by using some historical economics data (secondary data), evaluation of company specific performance by using secondary data like their shareholder annual report etc.  I also recalled those secondary data adopted need to be justified for its validity, reliability & recency.

Dear Eddie,

I think that everything helping to clear a scientific question or situation can be acceptable. So it is secondary analysis. I also used it earlier.

I used it in the paper below. Prof. Andras.  I am interested to know how other researchers do it. If you are willing, may I see a copy, at least of the methodology of your paper?

https://www.researchgate.net/publication/259822571_Relationship_Between_Gender_and_Employment_Contexts_among_the_Graduates_of_a_University?ev=prf_pub

Data Relationship Between Gender and Employment Contexts among th...

Secondary data analysis is considered scientific and an acceptable research approach. The source of the data should be authentic depending on the context of the research.

Good question.  The judgement about what constitutes good scholarly research tends to follow levels of evidence for varying types of research studies and analysis.  These level descriptions vary somewhat with different disciplines and colleges and universities.

Typically the lower levels are opinions of experts, then a singular qualitative or descriptive study, then evidence from systematic reviews of qualitative or descriptive studies, clinical trials that are quasi-experimental studies, evidence from randomized controlled trials (RCTs) and finally systematic reviews or meta-analysis of all relevant RCTs or multiple RCTs as the highest levels. 

What isn't as clearly defined but seems to follow similar levels is the  application of scholarly value by journal editors and reviewers and college and university committees for tenure and promotion.  RCTs generally are seen as more important than a qualitative or descriptive study. 

My experience with journal editors is they are more open to giving scholarly value to new ideas or directions of research even if they come from a qualitative or quantitative study.  If they pick up on a paradigm or framework shift that is significant regardless of the level they may be more enthusiastic about publishing that type of work.  Those of us who have qualitative research leanings most often believe the core concepts of most phenomenon should be first identified using qualitative studies before approaching quantitative studies. 

Looking forward to seeing your secondary data analysis in print!

Look at Emile Durkheim work on religiosity: The Elementary Forms of the Religious Life (1912). It refers mainly to aboriginal beliefs and is based solely on secondary sources (other people's studies), constituting desk research. He found a theoretical framing for the phenomenon, supported by empirical data from other scholars. So yes Emile Durkheim is a proof that secondary analysis can do the trick.

I found this document very useful, Anna.

http://home.ku.edu.tr/~mbaker/CSHS503/DurkheimReligiousLife.pdf

The assumption is that the experiences of the past will help to predict the future given that the rules do not express temporal change?

Dear Prof. Callaway,

Secondary analysis (the analysis and interpretation of raw data generated by someone else) and the related topic of secondary research (knowledge extraction from data analyses and interpretations performed by others) are becoming critical to progress in many fields. These secondary efforts have been enabled largely by rapid advances in various areas of Artificial Intelligence (AI).

The general field of AI has gone through several “AI winters”, wherein high initial hopes were quickly dashed when the true complexities of the problems that were focused on became all-to-apparent in applied efforts. However, machine learning, expert systems and other AI-derived tools have been applied with reasonable success to scientific (mostly numerical) data analysis for decades. Also, it is often the case that once a method derived from AI achieves mainstream success, it is no longer called AI. A good example here is automated speech translation.

The AI winters are over. I’m not going to attempt an extended review here; I just want to give an overview, and focus a little more on recent developments in legal research.

A prime example is in the field of cancer diagnosis and treatment. IBM and the American Cancer Society have teamed up to create “the ultimate cancer advisor” using Watson, IBM’s “cognitive computing platform”. Overall, this application could be considered an example of performance of secondary analyses that are simply beyond the capabilities of any individual clinician or medical researcher, or any medical research team using traditional approaches. See: http://fortune.com/2016/04/12/ibm-americancancersociety-partnership/

There are numerous other examples from other areas. For example, the National Science Foundation Project Geolink aims to produce a “smarter academic search engine” for geoscientists. J. You, Geoscientists aim to magnify specialized Web searching (2015) Science 347, 11. There are even efforts to automatic the process of scientific research itself. J. You, DARPA sets out to automate research (2015) Science 347, 465; Y. Gil et al., Amplify scientific discovery with artificial intelligence (2014) Science 346, 171-172.

AI-based methods are becoming increasingly important in legal research. The American Bar Association's new lawyer magazine discussed AI applied in law in an article "Law + Tech: Friend, Foe or Both" (http://www.americanbar.org/publications/tyl/topics/resources-technology/law-tech-friend-foe-both.html ). The article begins "What if you started typing a brief and, as you typed, case citations supporting your arguments magically appeared on the page? Sound far-fetched? Well, that’s just one of the projects of the “law+tech” movement, and it’s probably not long before that and more is possible." I could really use something like that!

The ability to mine unstructured text data (using natural language processing, computational linguistics and other advanced machine learning methods) has progressed to the point where a number of successful companies providing legal-domain AI services have emerged. The 2013 article "Supercharging Patent Lawyers with AI" (http://spectrum.ieee.org/geek-life/profiles/supercharging-patent-lawyers-with-ai ) first brought my attention to this field - it describes the development of the company Lex Machina (https://lexmachina.com/ ) out of a collaboration between the Law School and Computer Science Department at Stanford.

In addition to Lex Machina, other companies applying AI to legal research include IPVision ( http://www.ipvisioninc.com/ ), AcclaimIP (http://www.acclaimip.com/), ROSS Intellegence (http://www.rossintelligence.com/ ) and NeotaLogic (http://www.neotalogic.com/ ). Lex Machina (recently acquired by LexisNexis) is focused primarily on using AI methods to guide litigation strategies in IP Law and other areas of law. The IPVision platform is directed more towards patent claims, patent prosecution and portfolio analysis. Similarly, AcclaimIP provides very sophisticated patent analysis tools. The search terms can be entire sets of patent claims. NeotaLogic provides a "drag-and-drop" software platform that allows lawyers and other non-programmers to build custom-designed expert systems.

ROSS Intelligence is leveraging IBM’s Watson. As discussed on their site for lawyers (http://www.rossintelligence.com/lawyers ) "ROSS is an artificially intelligent attorney to help you power through legal research. ROSS improves upon existing alternatives by actually understanding your questions in natural sentences like - "Can a bankrupt company still conduct business?" ROSS then provides you an instant answer with citations and suggests highly topical readings from a variety of content sources." For reasons mentioned below, extending these approaches to other specialized areas (in the legal domain and across other fields) is confronted with a common, and very problematic, bottleneck.

A major limitation with most current AI methods for dealing with natural language documents is that they are founded primarily on “supervised learning” methods. Briefly, supervised machine learning encompasses a variety of algorithms that are "trained" using "features" extracted from a large number of "classified" objects. Generally, "classified" objects are ones that have been annotated, classified (and possibly corrected and/or re-organized) by human experts. Once sufficiently trained using the “labeled” objects, these algorithms can then correctly classify (in most cases) previously unclassified/unlabeled objects automatically. Natural language processing applied to text has reached the point where the semantic meaning (to varying degrees of correctness/completeness) of sentences, paragraphs and even complete discourses can be extracted. These extracted "semantic features" are used to train machine learning algorithms to classify previously unclassified text documents.

The limiting factor with both supervised machine learning (and also with expert systems approaches) is that a great deal of human effort is required up front to get something that actually provides meaningful results for a given domain. For machine learning, a great deal of data has to be "cleaned-up" and annotated before the algorithms can be trained sufficiently. In the case of expert systems, very large bodies of rules must be properly encoded.

It is likely that the human annotation bottleneck will eventually be overcome using developments from semi-supervised learning, another AI sub-field. Essentially, the idea is to start with a relatively small number of objects “labeled” by human experts and a relatively large number of unlabeled objects. The semi-supervised methods expand the number of correctly labeled objects until there are enough to confidently train supervised methods. In addition to numerical data sets, semi-supervised methods for annotation and classifying text documents have been described (http://www.google.com/patents/US7366705 ).

Once semi-supervised methods mature sufficiently, AI applications will expand very rapidly across numerous specialized domains.

I know this has been somewhat of a digression from the scope of your question, but I believe that secondary analysis and secondary research are scholarly strategies, both in their own right, and in their ability to inform decisions in traditional research. And in today’s data-driven world, secondary analysis and secondary research are also increasingly essential strategies.

Best regards,

Keith

I think the question has been answered already, and I can't say that my opinion is particularly well grounded (but it agrees with the optimistic inputs from the other contributors):

I can't see why it wouldn't be, so long as one is cautious about relevant stuff that might be inavailable to the researcher and influences the range of applicable methods and inferences from the dataset. Data is data, and the trust in ones colleagues already makes a fundamental precondition on any research, since we both cite peers, use taught background knowledge, and assume that contributers are truthful and so on. Secondary data is just data, perhaps with larger epistemic gaps as to sampling methods etc., but used correctly, it would appear arbitrary to rule it inadmissible as the foundation for original research. I don't know much about practices related to this, so take it as personal, but rather strongly felt, perhaps erroneous opinion.

Yes

Systematic reviews and metaanalysis are such studies and are valid scholarly work.Systematic reviews and metaanalysis of high quality RCTs constitute high level of evidence,but can not fully replace the need for a single large RCT or a multi centre RCT.

The preference for secondary/primary data primarily depends on the objectives of the research and of course, the availability of secondary data. Traditionally, some researchers or supervisors might accept primary data source as more authentic due to unavailability of valid secondary data. We need to appreciate that research based on secondary data is as rigorous and authentic as research based on primary data. Research supervisors should encourage the use of either source of data depending on the objectives of the research. I personally believe that research using secondary data is less prone to manipulation in comparison to primary data research. It needs to be noted that some of the highly acclaimed and widely cited research papers have employed secondary data source. There can be variations for the preference of the data source based on the field of research.