Making a Contribution in Accounting Research, Part II: Focus on Nonprofit Accounting
In a previous post I laid out the main different ways articles can make a “contribution” to the literature. In order to make my arguments more concrete, in this and subsequent posts I will be analyzing nonprofit accounting articles for theoretical “contribution.” Below you will find all 31 articles that have been published in five of the six “pinnacle” accounting journals as defined by inclusion in the Financial Times 50 list. Specifically, this includes The Accounting Review, Journal of Accounting Research, Journal of Accounting & Economics, Contemporary Accounting Research, and Review of Accounting Studies.
I have excluded the sixth journal, Accounting, Organizations and Society, for two reasons. One, there are a large number of “nonprofit” and “NGO” publications in this journal, which would complicate the goal here of a parsimonious analysis. Two, the great majority of the nonprofit publications in AOS are qualitative, while the nonprofit publications in the other 5 journals are exclusively quantitative; this also would complicate a comparison of intellectual “contribution.”
A few interesting notes: As shown in the References at the bottom of the post, there are 31 total publications, with 4 in RAST, 2 in JAR, 5 in JAE, 7 in CAR, and 13 in TAR.
In addition, as shown in the figure below, all but two of the publications occur after 2000. One in the 1970s in JAR, one in the 1990s in JAE, eight in the ‘aughts’ in JAE and TAR, and the remaining 21 all come after 2010.
In my subsequent post I begin analyzing the nature of the intellectual contribution of these 31 articles.
All Nonprofit Articles Published in TAR, JAR, JAE, and CAR
[bibtex file=nonprofit_articles.bib, format=apa template=av-bibtex-modified sort=year order=desc]
Making a Contribution in Accounting Research, Part I: Types of Contributions
This is first in a series of posts covering how to make an intellectual contribution in academic research in accounting as well as most other business and social science disciplines.
In the academic worlds I interact with — accounting, business, social sciences — one of the words you’ll hear time and again is “contribution.” This is particularly relevant when you receive a review of one of your own papers and read that it fails to make a large enough “contribution.” 🙂
So, what exactly constitutes an intellectual contribution in accounting research? The short answer is that it tells us something useful that we do not already know. That short answer is not too helpful, however. So, in this post I will convey my more concrete thoughts on what makes a relevant contribution in an academic article.
First, let me lay out some generic categories of contributions and which ones are seen as best in top-tier academic accounting journals. The same general lessons apply to quantitative research in any of the other social sciences or business disciplines (I’ll leave aside the notion of a contribution in qualitative research). The following is a distillation of ideas I’ve learned over the years. For a more detailed discussion of some of these points, see, among others Voss (2003), Slater and Gleason (2012), and Colquitt and George (2011). The main ideas are my take on things, however, so the usual caveat applies.
A first type of contribution is to replicate, or “re-do,” an existing study.
To help make these ideas concrete, I will use the specific example of what leads to earnings management in nonprofit organizations. Using an example of a strong, well-published accounting paper, let’s look at Eldenburg, Gunny, Hee, and Soderstrom (2011), who examined the relationship between pay-for-performance incentives (the independent variable, or IV) and earnings management (the dependendent variable, or DV) in US-based nonprofit hospitals. If you were to try to replicate this study, you would draw a new sample of US-based nonprofit hospitals, recreate the authors’ measures, and then re-run the statistics using the same statistical procedures as in Eldenburg et al. (2011). You will have thus replicated their study.
While replication studies are needed in academia, unfortunately, they are not highly valued by the pinnacle accounting journals (nor those in other business or social science disciplines). In fact, replication is the “lowest” type of contribution, while in practice means you would likely find your study published in a low-level (or if you’re lucky a mid-level) accounting journal.
New Sample or Context
Second, a contribution can be made by taking an existing relationship found in the literature and applying it in a new geographic or organizational context.
Returning to our Eldenburg et al. (2011) example, and pointing out that they based their empirical tests on US nonprofit hospitals, someone could come along and test the same IV → DV relationship in a different context. If a different geographic context, the relationship could be tested in a sample of Canadian, or Chinese, or Cambodian hospitals. If a different organizational context, someone else might test the IV → DV relationship in a set of US-based educational nonprofits or environmental nonprofits. In either case, the authors are asking, “We know that X → Y holds in US-based hospitals, but does it also hold in other countries or in other types of nonprofits?”
This is certainly a valid way to build our knowledge and helps us understand such things as idiosyncracies of distinct contexts as well as boundary conditions, etc. However, it is what I would call an incremental contribution. This type of contribution, as a general rule, will therefore be relegated to the mid-tier and lower-tier accounting journals.
New Measure (Often with Revised Conceptualization) or New Statistical Technique
A third type of contribution is to make a new measure of one of the key concepts in an established relationship. Continuing our example above, a researcher might come along and develop a new measure of earnings management and then re-test the relationship between pay-for-performance incentives and earnings management in US-based nonprofit hospitals. Often, but not always, this new measurement approach will be spurred by the author’s view that the definition of the concept needs updating.
I would also place in this category new statistical tests of the relationship between the independent and dependent variable. For example, if someone comes along and argues that the original test should be done with robust clustered standard errors, then that is an example of this type of argument.
Again, this is a valid way to enhance our knowledge. Nevertheless, if this is the chief contribution, it is another example of an incremental contribution that, as a general rule, will also for this reason be relegated to the mid-tier and lower-tier accounting journals.
Lastly, adding a new variable to an existing model— whether a mediator, moderator, or existing explanatory factor—is one of the fundamental ways of doing “innovative” social scientific research (Slater and Gleason 2012; Voss 2003). Note that in all of these cases, it is the additional of a new relationship (either new IV, new DV, or new IV/DV combination).
Here is the relationship:
Why is the relationship so important? The short answer: theory. Theory fundamentally involves examining the relationship between two or more concepts. We are making a theoretical argument when we try to explain why one concept (earnings management) is determined by another concept (pay-for-performance incentives).
Note that Eldenburg et al. did not come up with the idea that earnings management existed. Where they innovated was in positing and then testing the idea that earnings management was driven by pay-for-performance incentives (they also examined the effects of accounting performance, or the extent to which performance was above or below a benchmark).
In short, when people (reviewers, discussants, co-authors) are talking about a “contribution” or an “intellectual contribution,” they are almost always referring to a theoretical contribution. This means that at a minimum you should strive to add another conceptual relationship to the literature. This could be a new mediator, a new moderator, a new independent variable, or a new dependent variable. Of course, new definitions, new measures, and new contexts are also helpful but they should not be the sole novelty of any manuscript you’re thinking of sending to a top-tier journal.
[bibtex file=contribution.bib, format=apa template=av-bibtex-modified sort=year order=desc]
Making a Contribution in Accounting Research, Part III: Relationships in Top Nonprofit Accounting Articles
This post is Part 3 in my series on intellectual contribution in academic articles. In Part 1 I covered the main types of “contributions” at an abstract level. In Part 2 I turned to the first in a set of posts designed to make the typology concrete by analyzing the type of contribution in the 31 nonprofit-focused articles published to date in The Accounting Review, Journal of Accounting Research, Journal of Accounting & Economics, Contemporary Accounting Research, and Review of Accounting Studies.
The 31 articles are included at the end of this post.
Recap: Four Main Types of Contribution
Before I delve into specifics, let me recap the four central types of contributions:
- New Sample or Context
- New Measure (Often with Revised Conceptualization) or New Statistical Technique
- New Relationship
I argued that the first three contributions — replication studies and tests involving new contexts, new definitions, and new measures — are all helpful, but they should not be the sole novelty of any manuscript you’re thinking of sending to a top-tier journal. Instead, by far the most important contribution is testing a new relationship. Why are relationships so important? The short answer: theory. Theory fundamentally involves examining the relationship between two or more concepts. We are making a theoretical argument when we try to explain why one concept (say, earnings management) is determined by another concept (such as pay-for-performance incentives).In effect, when reviewers, discussants, co-authors, or thesis committee chairs are talking about a “contribution” or an “intellectual contribution,” they are almost always referring to a theoretical contribution, which effectively means you are studying one or more new relationships. This means that at a minimum you should strive to add another conceptual relationship to the literature. This could be a new mediator, a new moderator, a new independent variable, or a new dependent variable.
Analyzing the 31 Articles
So, what do I find in the 31 top-tier articles? I find that all 31 test a new relationship. None involve solely a replication, or testing an existing relationship in a new context, or testing an existing relationship with a new measure. In effect, all 31 aim for what I would consider the “gold standard” form of contribution.
Specifically, I found that the typical article examined 1-5 core relationships, not including robustness tests, sensitivity analyses, or additional analyses. Interestingly, few of the studies had core hypotheses with moderating relationships and instead looked at direct X → Y relationships.
Graphing the Relationships
The figure below maps the relationships found in the 31 articles. Click on the image to see a larger version.
You can see an interactive version of the graph here.
In Part 4 I present the specific relationships examined in each article and show the Networkx code used to generate the network and the above visualization.
List of 31 Articles Published in Top-Tier Accounting Journals
[bibtex file=nonprofit_articles.bib, format=apa template=av-bibtex-modified sort=year order=desc]
Quest for Attention: Nonprofit Advocacy in a Social Media Age
The past year and a half I’ve been working on an academic book for Stanford University Press with my friend Chao Guo, Associate Professor of Nonprofit Management in the School of Social Policy & Practice at the University of Pennsylvania. We’re happy to say, the book is out!
|Guo, Chao, & Saxton, Gregory D. (2020). Quest for Attention: Nonprofit Advocacy in a Social Media Age. Stanford, CA: Stanford University Press.|
The book is a culmination of the work Chao and I have done that draws on our understanding of nonprofit organizations, advocacy efforts, and data analytics. Here’s a brief description of the book:
Today, social media offers an alternative broadcast and communication medium for nonprofit advocacy organizations. At the same time, social media ushers in a “noisy” information era that renders it more difficult for nonprofits to make their voices heard. This book seeks to unpack the prevalence, mechanisms, and ramifications of a new model for nonprofit advocacy in a social media age. The keyword for this new model is attention. Advocacy always starts with attention: when an organization speaks out on a cause, it must ensure that it has an audience and that its voice is heard by that audience; it must ensure that current and potential supporters are paying attention to what it has to say before expecting more tangible outcomes. Yet the organization must also ensure that advocacy does not end with attention: attention should serve as a springboard to something greater. We elaborate how attention fits into contemporary organizations’ advocacy work and explain the key features of social media that are driving the quest for attention. Developing conceptual models, we explain why some organizations and messages gain attention while others do not. Lastly, the book explores how organizations are weaving together online and offline efforts to deliver strategic advocacy outcomes.
On this page I will be placing tutorials, additional analyses, and replication code related to the book.
Please check back again soon!
Chapter 3 – Analyzing Twitter Data by Time Period
Chapter 2 – Aggregating and Analyzing Data by Twitter Account
What is Accounting Analytics?
One of the buzz-words in business schools is data analytics or, in an accounting school, accounting analytics. But what exactly is ‘accounting analytics’? How is it different from existing tools and disciplines such as ‘statistics’, ‘computer science’, ‘machine learning, ‘Big Data’, or ‘managerial accounting’? In this post I will disentangle this emerging field. This is a first crack at this issue — I will continue to edit as the field (and my understanding of it) develops.
Accounting Analytics Defined
Accounting analytics, in a nutshell, is the examination of Big Data using data science or data analytics tools to help answer accounting-related questions. Let’s break that down further.
The Emergence of Big Data
The first piece of the puzzle is `Big Data.’ Big Data is different from previous forms of data in terms of volume, variety, and velocity: There is a lot more of it, it comes in a variety of structures and formats, and it is updated and produced quickly. Any of the following might be considered sources of Big Data for an accounting analytics project:
- social media data
- web search data
- journal entries
- transactional data (e.g., customer transactions)
- call centre transcripts
- store videos, web cams, etc.
- online customer reviews
- proprietary databases
- IoT devices
- image repositories
- sensor data
- public data (crime, health, education, etc.)
- weather forecasts
- company emails
- company disclosures
There are lots of good articles and blog posts out there on this point. The key to recognize is that Big Data goes beyond just social media data.
The Emergence of Data Science
The second piece of the puzzle is `Data Science.’ Data scientists examine Big Data using a combination of programming skills, statistical skills, and domain knowledge to answer relevant organizational or societal questions.
Data science has emerged in the past decade with the emergence of Big Data and the concurrent development of readily accessible sophisticated computing tools. At its core, data science is a field, or perhaps an approach, that lies at the intersection of three things: 1) computer programming, or ‘hacking’, skills; 2) mathematical and (especially) statistical skills; and 3) domain expertise.
Programming skills are needed to help access and download Big Data, to `wrangle’ the data into a usable state, and to write programs that can run sophisticated machine learning and related algorithms. Statistical skills are needed to ensure the appropriate research design methodologies are followed and statistical techniques applied; without this, your projects findings will not have validity. Finally, domain expertise is needed in order to identify questions that are worth answering. In accounting analytics, domain expertise is knowledge about the business and industry along with understanding of the accounting information system.
Data Analytics vs. Data Science
There does not appear to be a consensus on the distinction between ‘data science’ and ‘data analytics.’ You can refer to a number of articles and blog posts that attempt to describe the difference. For example, some think of data analytics as a narrower sub-field of data science. To my mind, however, the two terms are more or less synonymous. Accordingly, accounting analytics can be thought of as the use of data science/data analytics techniques to answer accounting-related questions.
Generic Data Analytics Tools
A well-rounded data scientist needs to have a broad ranges of tools and techniques in her toolkit. We might think of these as encompassing four stages of the process: 1) data gathering, 2) data wrangling, 3) data management, and 4) data analysis.
Hacking skills are indispensable for accessing and downloading certain types of Big Data — particularly social media data. Here the data scientist will need to become familiar with application programming interfaces (APIs). I have written a blog post on how to set up access to the Twitter API, and there are lots of other resources out there.
Data wrangling or ‘data munging’ is a critical data analytics skill. Once you have the data you then need to get it into a usable format. This might involve generating ‘pivot tables’ in Excel or PANDAS or R, or aggregating a time-series dataset to the daily or weekly or monthly level, or generating a series of new variables (aka ‘feature engineering’). And it might involve all of these things. Big Data are often unstructured data, meaning they are lacking pre-defined fields (aka ‘columns’ or ‘variables’) and linkages across databases. I don’t want to devote too much time to this issue as you’ll find plenty of resources online. The key takeaway is that data wrangling is something any social scientist or other type of researcher is familiar with — with Big Data there are just additional challenges in terms of scope and nature of the issues you’ll be dealing with.
Data scientists should have familiarity with database management. There are lots of options out there, ranging from traditional relational databases such as SQL to ‘noSQL’ databases such as MongoDB. In my own research I have recently leaned toward the MongoDB approach, but the data scientist may not have a choice. For this reason a solid understanding of both major approaches is helpful.
Once you have the data in usable format and you’ve found a question worth answering, you then need to analyze the data. There is a huge range of techniques for analyzing the data. In my training as a social scientist and then accounting scholar, I was trained to think of two broad sets of techniques — descriptive statistics and inferential statistics. Examples of the former are the mean, mode, median, and standard deviation, and are used to describe the data that you’re working with. To help deliver answers to a given research question, we then turn to a wide range of more advanced statistical techniques, including ordinary least squares (OLS) regression, logistic regression, negative binomial regression, etc.
Data scientists avail themselves of all these techniques in addition to a much broader range of tools. This is due in part to the rapidly developing field of computer science as well as the different types of questions data scientists are asking (it is a long story, but the crux of the argument is ‘prediction vs. explanation’).
Here the data scientist will be familiar with all of the ‘buzzwords’ — artificial intelligence, deep learning, machine learning, text mining, sentiment analysis, clustering, and more. Again, I don’t want to reinvent the wheel, so any number of articles out there can explain these tools and how they are applied in data analytics.
OK, by this point I hope you have a better sense of what ‘data science’ is. What you may not see is how this is practically applied in the field of accounting analytics. Here I have a somewhat different take than other articles and blog posts I’ve read. Most of them seem to categorize data analytics and/or accounting analytics questions into a ‘descriptive’ (what is happening) vs. ‘retrospective’ (backward-looking) vs. ‘prospective’ (what will happen) framework. While accounting analytics is certainly used for description, retrospection, and prediction, I prefer to approach this field differently. Namely, I focus on two main research questions: 1) measurement and 2) finding relationships.
Measuring Accounting Concepts
A first set of accounting analytics questions are designed to help measure an accounting-related concept. Accounting is inherently a ‘measurement discipline’, but accounting information systems are not equipped to deliver insights into all variables of interest. Here’s a concrete example. Let’s take an intangible asset such as a company’s brand community. The company has built this intangible asset itself so it is off the books — it will not be seen on the balance sheet and thus will not automatically be measured.
As shown in the figure below, it is critical that the accounting analytics practitioner first recognize that the notion of ‘brand community’ exists at the conceptual level — we have no solid concrete measure of it. Instead, we need to examine a range of different indicators that, collectively, can help us measure this concept. The figure below shows four variables, or indicators, that could be examined to help infer the strength of the company’s brand community.
None of these indicators in isolation would suffice. It is only by ‘triangulating’ our findings that we are able to make any valid inferences. And it is only by employing data analytic techniques that we are able to gather and wrangle these data.
Most accounting students are not trained in separating the conceptual from the measurement level, yet it is a critical piece of the puzzle. It is one of the core competencies of the data scientist’s training in statistics and methods.
The second type of accounting analytics question addressed is the search for relationships between one variable and another. For example, the accounting analytics practitioner might be interested in what behaviors predict fraud. This is the search for the relationship between some set of ‘X’ behaviors and a ‘Y’ outcome variable (fraud). This is an ‘accounting’ analytics question because of the accounting-related variable fraud.
Similarly, the analytics practitioner might wish to segment job applicants into ‘good’ and ‘bad’ applicants. The analytical tool might involve a clustering algorithm, and as in the above example, ‘under the hood’ the analyst is using a variety of variables help separate promising from less promising applicants. In other words, the exercise involves finding patterns built concerning the relationship between a bunch of “X’s” and an outcome variable “Y” (job applicant quality).
I’ll give one last example here. The figure below shows an accounting-related variable sales. Sales data would already be gathered in the accounting information system. Here is where many accountants would stop. Accounting analytics, however, brings something new to the table. Namely, accounting analytics is not only interested in measuring sales but in relating sales to other variables. In other words, the accounting analytics practitioner is keenly interested in seeing what other variables can predict the increase (or decline) in sales. This is the search for relationships. Where data science enters the picture is in two ways. One, inferring relationships is inherently a statistical or methodological pursuit, and training in statistics is not typically a forte of the accountant. Two, the data employed is often non-financial in nature. Big Data is commonly used and thus data science tools are a necessity.
As shown in the figure, the accounting analytics team is interested in predicting future sales for the company. To help answer this question, the team gathers an array of financial data (not shown) as well as non-financial data, including the strength of the online brand community, the number of five-star Amazon reviews, an assessment of the company’s CSR and sustainability initiatives, weather patterns, and customer complaints. Once these data are gathered and wrangled, operational measures are developed and then statistical analyses are applied. Based on these analyses the team will have answers to what the strongest predictors of sales performance are.
These are just a few examples. There is a huge range of questions that could be answered using accounting analytics. The key is these questions involve at least one ‘accounting’ variable and rely on Big Data and data analytics tools.
What The Typical Accountant Needs to Know
The great majority of accountants (and accounting students) are not going to become data scientists. What all practitioners should learn is data literacy. They need to know understand what types of questions can be asked (measurement and finding relationships), what types of information are available (Big Data), and what types of analyses can be run. In short, the typical accountant does not need to know how to do data science but does need to understand what data science is. They should, in Cornelissen’s (2018) words, “speak the language of data.” Being ‘data literate’ will help you and your team make better requests of the data scientists. As seen in the Henke et al. (2018) article linked below, a recent buzzword points to a new career path for those who are literate in data analytics — the ‘analytics translator.’
This is intended to be a relatively brief overview of the rapidly emerging field of accounting analytics. I hope this article has provided a useful overview.
Accounting academics and practitioners are rapidly building knowledge of this emerging field. In an update I will add some key academic articles for further reading. For now, I refer you to four brief Harvard Business Review articles.
- Andrew McAfee and Erik Brynjolfsson. Big Data: The Management Revolution, Harvard Business Review, October 2012.
- Hugo Bowne-Anderson. What Data Scientists Really Do, According to 35 Data Scientists., Harvard Business Review, August 15, 2018.
- Jonathan Cornelissen. The Democratization of Data Science., Harvard Business Review, July 27, 2018.
- Nicolaus Henke, Jordan Levine, and Paul McInerney. You Don’t Have to Be a Data Scientist to Fill This Must-Have Analytics Role., Harvard Business Review, February 5, 2018. (Also available here)