2.1 The Corpus

The corpus is drawn from journals categorised as public policy and administration journals by Google Scholar Metrics and those in the public administration sections of the Journal Citation Reports published by Clarivate and Scopus’ Scimargo Journal Rankings for 2016 (the date we commenced work on this project). The number of journals included in these indexes varies, for example, the Scimargo Journal Rankings is the most comprehensive, including all journals in the other indices and providing a population of over 100 journals, while there are only 47 journals listed in the Journal Citation Reports. Not all journals in these listings focus exclusively on public administration. For example, Climate Change and Policy Sciences are listed in the public administration category in the Journal Citation Reports, the Scimargo Journal Rankings includes Administrative Science Quarterly and Educational Administration Quarterly, while Google Scholar Metrics includes Science and Public Policy. Some public administration journals are highly specialised by topic and geography, for example Public Budgeting and Finance, the Journal of Homeland Security and Emergency Management and the Transylvanian Review of Administrative Sciences.

To resolve these variations, we implemented a set of decision rules to ensure that our corpus consisted of journals that have the public administration discipline as their core area of enquiry. First, we specified that the primary focus of the journal should be public administration – for example, while Scimargo lists Administrative Science Quarterly as the top journal in its public administration category, it publishes a limited number of public administration studies. Its primary focus is organisational theory in different settings. Second, selected journals had to be listed in Google Scholar Metrics, the Journal Citation Reports and the top 50% of Scimargo Journal Rankings (to ensure high levels of rigor of the published articles). The third and fourth criteria were the exclusion of specialist and non-English language journals, respectively.

The application of these decision rules resulted in a corpus consisting of the seventeen journals listed in Table 1, which also shows the year each journal was first listed in the Journal Citation Reports. Journal articles from the 1991–2019 study period were scraped from Web of Science for their metadata, including title, abstract, author information and publication year. The final sample contained 12,760 articles.

Limiting the scope of the study was necessary to ensure its feasibility, and confining the sample to journal articles ensured that the studies met the rigorous standards of scholarly publications. Although this excluded books and reports from governments and international organisations such as the OECD, bias arising from a focus on journal articles has been shown to be small (Rosenthal Reference Rosenthal1991).

2.2 Topic Modelling

Topic modelling uses a natural language processing technique called Latent Dirichlet Allocation (LDA) to extract topics. Latent Dirichlet Allocation was developed by computer scientists in the mid-2000s and has been extended beyond the field of computer science. The best way to understand topic modelling is the article on probabilistic topic models by Blei (Reference Blei2012). Blei (Reference Blei2012) showed how using optical character recognition to scan the entire collection of Science from 1990 to 2000, a sample containing 17,000 documents and 11 million words, including 20,000 unique terms or words (after excluding stop words and rare words), could produce an LDA output of 100 topics. Advances in this method have explored different classes of topic modelling such as dynamic and correlated topic modelling, structural topic modelling and supervised topic modelling, among others.

Topic modelling is an unsupervised machine learning technique that has gained popularity in other disciplines in the social sciences (e.g., political science, communication studies, management and organisations), although it has only recently been applied to the field of public administration. For example, Chandra et al. (Reference Chandra, Jiang and Wang2016), Walker et al. (Reference Walker, Chandra, Zhang and van Witteloosijin2019), Li et al. (Reference Li, Chandra and Fan2022) and were among the earliest researchers to use LDA in public administration to examine cross-sectoral governance. Chandra et al. (Reference Chandra, Jiang and Wang2016) used LDA to investigate the patterns in the strategies used by social entrepreneurs to bring about social change. These authors relied on theories from the field of social entrepreneurship to classify the change-making strategies. Walker et al. (Reference Walker, Chandra, Zhang and van Witteloosijin2019, Reference 97Walker, Zhang, Chandra, Dong and Wang2023) used LDA to conduct a bibliometric study of the evolution of the research–practice gap, a long-standing debate in the field of public administration. Li et al. (Reference Li, Chandra and Fan2022) used LDA to examine the social media messaging strategies adopted by different levels of the Chinese government during the Covid-19 pandemic. The authors relied on theories from the field of communication to interpret the patterns in the topics that were produced by LDA. Latent Dirichlet Allocation has also been applied to understand patent data (Blei and Lafferty Reference Blei and Lafferty2007; Kaplan and Vakili Reference Kaplan and Vakili2014).

There are good reasons why topic modelling is useful for social scientists. First, data in the social sciences are often in the form of text. Although some scholars in the social sciences primarily use survey or secondary (closed form) data in their research, many scholars work with textual data. Textual data are interesting because they are unstructured, messy, rich and thick, but can be confusing to read and synthesise, even for well-trained and experienced scholars. This speaks to the importance (and limitations) of human visual acuity in the understanding and detection of ‘hidden patterns’ in texts. Second, researchers in the social sciences who rely primarily on textual data tend to work with a small body of texts. For example, scholars who primarily work with qualitative data (e.g., in-depth interviews, ethnographic data, newspapers) can only analyse a small number of such texts. It is difficult to imagine how a researcher could manually code and synthesise thousands or hundreds of thousands of pages of texts and come up with reasonable results without being accused of ‘cherry picking’. These limitations motivate the use of a computerised tool such as the LDA as a method that can reveal the hidden patterns in a large body of textual data by using iterative cycles of analysis until patterns can be discerned.

To non-experts, LDA can be understood using the analogy of traditional content analysis: the purpose of LDA is nothing more than to synthesise the content of a body of textual data, just like traditional content analysis does. While content analysis can be done manually by human coders and typically qualitatively (e.g., texts such as newspaper articles are coded without the use of predetermined categories), it can also be done quantitatively (e.g., using a coding scheme to assign numerical values to a sample of texts treated as data and then conducting statistical analysis of these numerical values). To those who are more conversant with general linear models in frequentist statistics, LDA can be seen as another data reduction method similar to exploratory factor analysis (EFA). However, while EFA requires close-ended data (e.g., 1–5 or metric scales), LDA is applied to textual data. However, both methods have the same purpose: to reduce the dimensions to a manageable number so that the results can be easily interpreted, more parsimoniously. There are some trade-offs in how to decide which outputs to use. Latent Dirichlet Allocation is more positivistic (versus interpretative) in its epistemology, from the perspective of the philosophy of science.

Latent Dirichlet Allocation works by making inferences from observed meanings (or ‘topics’) in a large body of textual data, for example books, journal articles, newspaper articles, magazines or patent data (Blei Reference Blei2012; Blei and Jordan Reference Blei and Jordan2003). This means that LDA is probabilistic in nature and thus is not an absolute or deterministic method with closed-form mathematical equations as outputs. (Although the equations driving the algorithms in LDA in Python or R programming languages are nothing short of fantastic mathematical equations.) The probabilistic nature of LDA also means that for the LDA method to work effectively, it requires human intervention – humans make decisions on how many topics to produce, examine, interpret and provide labels to the topic outputs and reiterate the process until the best outputs are achieved. The word ‘best’ here usually means that the topics are easy to discern, are clearly differentiated and can be ranked according to their importance (e.g., by topic weight, semantic coherence, exclusivity).

Latent Dirichlet Allocation is therefore essentially a collection of computer algorithms. Technically, LDA assumes that each document (e.g., a Harry Potter novel) contains a variety of topics, but that the topics are usually unobservable or latent (e.g., Harry had a bad childhood is one topic; Harry, Ron and Hermione’s relationship is another topic), and these topics stand between (i.e., do not appear vividly to readers because they are ‘hidden’ in an ocean of words) the documents and terms (Blei Reference Blei2012; Paul and Dredze Reference Paul and Dredze2014; Tirunillai and Tellis Reference Tirunillai and Tellis2014). This essentially means that topics are meso-level information that are not easily spotted by human readers, while the actual words (or terms) and documents are the highly observable micro and macro levels of texts, respectively.

By laying out the terms (or observable ‘micro’ level data listed in rows) that exist in a set of documents (or observable ‘macro’ level data listed in columns) in a dataset (a matrix that can be produced using any computational tools), LDA probabilistically infers what topics (or ‘meso’ level information) exist in the documents (Chandra et al. Reference Chandra, Jiang and Wang2016; Landauer et al. Reference Landauer, McNamara, Dennis and Kintsch2013). Far from being a straightforward, one-way inference, LDA works iteratively by first making assumptions about the relationships among terms and the documents and then making inferences from the posterior expectation or posterior inference problem (that is, the p (topics, proportions, assignments | documents)). Specifically, it assigns topics and proportions of topics to words and documents where (i) each topic is associated with a distribution of different words, (ii) each document contains a mixture of various topics and (iii) each word is drawn from one of the topics (see Blei Reference Blei2012). Latent Dirichlet Allocation achieves its optimal results using trade-off rules. That is, for each document, LDA allocates the words or terms to as few topics as possible (e.g., Document 1 is 95% Topic A, 3% Topic B and 1% Topic C, whereas Document 2 is 20% Topic A and 80% Topic B). For each topic, it assigns a high probability to as few terms as possible (e.g., two common topics in news stories are ‘inflation’ and ‘politics’ and the most common terms associated with inflation might be ‘price’, ‘high’, ‘shortage’ and ‘war’, while the words associated with the politics topic might be ‘NATO’, ‘Russia’, ‘Ukraine’ and ‘war’; note that the term ‘war’ is common in both topics). Achieving a high probability that a term represents a certain topic and reducing the number of terms used in the analysis are contradictory goals – requiring a trade-off – because the former requires assigning all of the words in a document to a few topics, while the latter requires (grouping) multiple words to have a high probability of being used in a single topic. The result of this trade-off is the identification of tightly co-occurring words. Accordingly, LDA is a general tool rather than a tool for a specific purpose. Humans can observe the documents and the terms, but the hidden structures, which are the topics, are not visible to human reader because of the vast number of words, and LDA uses algorithms to infer the topics and make them visible based on the distribution of words in the documents.

Accordingly, LDA uses the distribution of terms over documents (or a ‘document–term matrix’) to probabilistically estimate term–topic relationships (or a ‘term–topic’ matrix) and document–topic relationships (or a ‘document–term matrix’). In other words, LDA uses a number of ‘terms’ (e.g., sushi, burger, curry; summer, airline, hotel) to classify and define ‘topics’ (e.g., ‘food’, ‘holiday’); each document can be associated with several topics (e.g., food, Japan, well-being), but each document has a primary topic represented by the highest topic–document probability with lower probabilities for other topics.

Among the most popular techniques for implementing LDA is the topicmodels package for the R programming language (Hornik and Grün Reference Hornik and Grün2011). Other popular R packages for topic modelling include the tm package (Feinerer Reference Feinerer2015), which is often used to pre-process and convert the corpus to create a document–term matrix. Other packages include dplyr (for unstructured data manipulation), quanteda (a general tool for natural language processing, including data manipulation and sentiment analysis) and ggplot2 (for data visualisation). One can also perform LDA using other programming languages such as Python with the help of pyLDAvis, Gensim or nltk pandas.

The first step in our LDA methodology was to capture journal names and author information (e.g., names, country of origin, department and university affiliation). We primarily relied on ‘title + abstract’ as the main data for topic modelling analysis. We used the topicmodels package in R to estimate and identify an initial list of topics (e.g., representative bureaucracy, collaboration, performance management) in the sample corpus of journal articles. We followed the best practices for LDA (e.g., Chandra et al. Reference Chandra, Jiang and Wang2016; Quinn et al. Reference Quinn, Monroe, Colaresi, Crespin and Radev2010) by involving human experts who read and interpreted the LDA topic outputs and provided names for each topic, that is, ‘topic labels’ (e.g., the group of terms ‘learning, improvement and measurement’ was labelled the ‘performance management’ topic). Any obviously overlapping topics were merged (relying on expert readers’ consensus) to ensure a more meaningful analysis. As LDA is a probabilistic tool for discovery (i.e., a bottom-up approach), we did not predetermine the number of topics at the outset but rather worked with the topics interactively, interpreting them, labelling them and then narrowing them down to a manageable number of topics.

Following the identification of key topics, we calculated the ‘total topic weight’ for each topic identified in the corpus and assigned a primary topic to each journal article. The total topic weight was a proxy for the relative popularity of a topic. The total topic weight was the dependent variable. This analysis was undertaken on the whole corpus and then separately for each journal.

2.3 Corpus Linguistics

We complemented the topic analysis with computational linguistics, or corpus linguistics, which allowed us to further analyse the characteristics of the corpus. Corpus linguistics is a discipline that examines patterns of language using statistical methods where a target corpus is compared with a reference corpus; it is also a methodology of choice of scholars in various disciplines from political science, marketing, digital humanities to terrorism and media studies. Using WordSmith, a popular corpus linguistics software, we studied the keyness or aboutness of the corpus by calculating the LLR of time at the discipline and journal levels by dividing the sample into early and late periods and comparing the topic distribution in the two subsamples. Here, we first explain corpus linguistics.

Corpus linguistics is a method for analysing the language in large bodies of text (Chandra Reference Chandra2016a, Reference Chandrab; Rayson Reference Rayson2008). A body of text is known as a corpus in the singular and corpora in the plural. In essence, corpus linguistics is a methodology for analysing language data in either spoken or written form (although most spoken data are transcribed into text form for corpus linguistics analysis) using corpus linguistics software (Rayson Reference Rayson2008). Corpus linguistics has a qualitative heritage; it existed in manual form for decades prior to the age of computerisation and this form continues to be used for studies with small samples (Chandra and Shang Reference Chandra and Shang2019) that are analysed by ‘hand and by eye’ (McEnery and Hardie Reference McEnery and Hardie2012, p. 3). However, in today’s digital age, scholars who conduct analysis of large corpora (Perren and Sapsed Reference Perren and Sapsed2013) usually perform the analysis using sophisticated corpus linguistics machines. However, the affordance of corpus linguistics has been greatly enhanced by the digitisation of text, the public availability of materials on the Internet (Bennett Reference Bennett2015) and the rise of modern computational tools that support corpus linguistics analysis such as WordSmith, BNCWeb, AntConc and the open-source package in R (Gries Reference Gries2016). These data and technologies are external enablers that allow researchers to manipulate and process corpora make up of millions and potentially billions of words, ushering in an era of digital humanities or digital social sciences. Accordingly, advances in computing technologies have sparked a ‘quantitative turn’ in corpus linguistics.

The texts used in corpus linguistics are usually gathered using systematic sampling techniques focused on a specific research question, literature or theory (and at times on certain hypotheses), organised into datasets (the corpora). These are then explored or validated quantitatively and qualitatively (McEnery and Hardie Reference McEnery and Hardie2012). The data can also be drawn from a census for some research question (e.g., studying how words co-occur or the aboutness of all articles published in Nature or Science). The findings in corpus linguistics are presented in a variety of formats, including tables that show the ratios of the frequencies of the focal objects in the target corpus against a reference corpus; mathematical formulae (e.g., measures of association of words such as mutual information); strings of words (e.g., n-gram analysis); statistical figures (e.g., LLR, chi-square statistics of the chance of occurrence of words); in visual representations (e.g., semantic category analysis showing categories of words with different probabilities of occurrence using larger or smaller font size for ease of viewing) and concordance analysis (showing a keyword and one word to the left and one word to the right of a keyword of interest, and agglomerative clustering of words, and analysing them qualitatively to discover themes). No single study uses all of these techniques; the selection depends on the researchers’ objectives and research questions.

Researchers interested in the association of a particular word (e.g., the corpus linguistics study of red tape by Kaufmann and Haans (Reference Kaufmann and Haans2021)) have focused on bi-gram analysis. While unigram is essentially a single word, such as ‘coffee’, ‘football’ or ‘email’, bi-gram, or lexical bundle, means two words that are frequently used together, such as ‘drink coffee’ or ‘play football’ or ‘send emails’. But bi-gram can also take different forms. For example, if one takes all published papers on Covid-19 and analyses their titles, the bi-grams could include ‘Covid-19 and’, ‘Covid-19 pandemic’, ‘of Covid-19’, ‘the Covid-19’; these can be used to understand the language patterns use but also the attention of researchers to the topic. When we pair three words together as a bundle, this is called tri-gram (e.g., ‘the Covid-19 pandemic’, ‘Covid-19 vaccination is’ and ‘Covid-19 caused many’; and when four words are paired together, this refers to four-gram (e.g., ‘the Covid-19 pandemic has’, ‘Covid-19 vaccination is administered’ and ‘Covid-19 caused many deaths’). The list can go on into what we call n-gram. Thus, n-gram is a short phrase of a variety of word pairs that are being analysed. N-gram is useful because researchers are interested in understanding how language producers (e.g., policymakers, learners, terrorists, surgeons and politicians) deploy language in a particular area.

Bi-grams are parsimonious and are useful when a researcher is interested in understanding how policymakers deploy language in a particular policy area. To understand language use – and how it fits into the arena of public administration – researchers may analyse the parliamentary (or congressional) discourse, media discourse or campaign language used by certain policymakers. Scholars have applied corpus linguistics to an archival corpus of parliamentary records in Westminster states to examine how they differ across the United Kingdom, Canada, Singapore and Vanuatu or to a specialised but small corpus from a particular organisation (Perren and Sapsed Reference Perren and Sapsed2013; Rheault and Cochrane Reference Rheault and Cochrane2020). For example, Perren and Sapsed (Reference Perren and Sapsed2013) analysed British parliamentary records in relation to innovation policy over a forty-five-year period and revealed that the term ‘innovation’ has gradually gained political importance, both in absolute numbers (i.e., frequency of use) and qualitatively, as shown by its position within political discourse (and by extension government policy and action). As another example, Chandra (Reference Chandra2016a) used corpus linguistics to understand the language of change-making used by social entrepreneurs (defined as individuals who combine business and nonprofit logics to tackle societal problems such as poverty, discrimination or climate change). Chandra performed collocations (bi-gram analysis) on a social enterprise taxi that serves disabled and elderly passengers using a small, specialised corpus, and then selected several keywords and drilled deeper using concordance analysis. He showed that corpus linguistics can serve as a tool to assist scholars to build and/or test theories. For example, he found that ‘wheelchair’ collocates with ‘user’, ‘bound’, ‘accessible’ and ‘friendly’; and ‘disabled’ collocates with ‘persons’, ‘permanently’, ‘carry’ and ‘transportation’; while ‘elderly’ collocates with ‘homes’, ‘commission’ and ‘wheelchair’. These allow the researcher to understand how such terms are used by the social entrepreneur of interest (in the interviews) and media coverage about the social enterprise concerned.

Corpus linguistics underlines the idea that ‘words matter’ (Perren and Sapsed Reference Perren and Sapsed2013, p. 1726) in our understanding of and practice in the social world, particularly for public administration. The field of public administration is inherently linguistic; it is expressed and performed through language. This understanding means that public administration scholarship should analyse the ‘talk and text’ (Prior et al. Reference Prior, Hughes and Peckham2012, p. 272) of public administrators as well as the works of public administration scholars. This summarises succinctly the main objectives of corpus linguistics: to complement, triangulate and provide empirical rigor to the conventional positivist mode of research (e.g., surveys, experiments) and constructivist field discourse analysis (Fairclough Reference Fairclough2013). Corpus linguistics offers a hybrid method associated with both positivism and constructionism.

Far from being dominated by a single field, corpus linguistics is interdisciplinary in its application. Researchers have used corpus linguistics to study linguistics, communication, law, sociology and management studies (O’Reilly and Reed Reference O’Reilly and Reed2011), as well as political science (L’Hôte Reference L’Hôte2010), public health (Prior et al. Reference Prior, Hughes and Peckham2012) and medical science (Pastrana et al. Reference Pastrana, Jünger, Ostgathe, Elsner and Radbruch2008). Researchers have used corpus linguistics techniques to examine questions related to actors (government and non-government, business and nonprofit), objects (public administration, service delivery and cross-sectoral partnerships) and organisations (e.g., social enterprises; Chandra Reference Chandra2016a; Parkinson and Howorth Reference Parkinson and Howorth2008). However, corpus linguistics remains rarely used in public administration. (An exception is Kaufmann and Haans’ study (Reference Kaufmann and Haans2021) of red tape.)

Viewing corpus linguistics (Baker and McEnery Reference Baker and McEnery2005; Chandra Reference Chandra2016a, Reference Chandra2016b; Rayson Reference Rayson2008) as a natural complement to topic modelling analysis, we used it in combination with our LDA analyses. We used the WordSmith tool (Scott Reference Scott2008) to perform corpus linguistics analysis and to pre-process the corpus into a suitable format for analysis. We divided the corpus into several sub-corpora (e.g., early versus late periods) and focused on the following techniques. First, we examined the keyness (the aboutness) of the corpus by calculating the LLR of a target corpus or sub-corpus (e.g., the early period) relative to a reference sub-corpus (e.g., late period). This produced a list of terms or words with high to low LLRs and their p-values across the target and reference corpus.

2.4 Geography of the Topics in Public Administration

To undertake a comparative analysis of topics across countries/regions, we grouped the topics according to the country/region the first/corresponding author was located in for each article. This approach may face limitations such as the first or corresponding authors being based in a different country than those examined in the article. However, this variable does capture the geography of scholarship in public administration by identifying the country of the university/institution showing leadership and the generation of knowledge in this field.

To provide a framework to understand the geography of scholarship, we used the World Bank Governance Indicators. These indicators evaluate various dimensions of governance across 213 countries and territories for the 1996–2020 period. We categorised the countries/regions in our sample using these indicators. The time period for the World Bank Governance Indicators overlaps with our sample period, permitting analysis. Six dimensions of governance are included in the index, and each dimension is measured on a scale from –2.5 to +2.5, with higher values indicating better governance. While the indicators may have their inherent biases or limitations and are subject to interpretation and criticism, they provide a valuable tool for understanding governance trends and informing policy discussions worldwide. The indicators are as follows.

• Voice and accountability: The extent to which citizens can participate in the selection of their government, freedom of expression and the existence of free media.

• Political stability and absence of violence: The likelihood of political instability, violence or terrorism within a country.

• Government effectiveness: The quality of public services, the competence of civil servants and the government’s capacity to implement policies effectively.

• Rule of law: The extent to which the government and its agents operate within a framework of laws and regulations, and the degree to which the judiciary is independent and impartial.

• Regulatory quality: The ability of the government to formulate and implement sound policies and regulations that promote private sector development and protect the public interest.

• Control of corruption: The presence of corruption in the public sector and the effectiveness of anticorruption efforts.

Our method had two steps. First, the three authors independently evaluated the conceptual overlap between the fifty public administration topics identified in the LDA analysis and the governance indicators and assigned each topic to one governance indicator. In the first round, inter-coder reliability (Krippendorff’s alpha) was 0.63. During the second round, we conducted extensive discussions and reached on an agreement on the topic assignments. The results of this analysis are presented in Section 5.

Second, to measure geographic differences, each country/territory was given a percentile governance rank based on the World Bank Governance Indicators scheme, ranging from 0 (lowest) to 100 (highest). We averaged the percentile rank data from 1996 to 2020 and calculated the score for each governance dimension. We further divided the samples into low and high score subsamples, with 80 as the cut-off value for the high score subsample. The rationale was based on the distribution of the data: the articles in our sample had an average score of 80–90 on the six indicators with a standard deviation of approximately 10. Therefore, if a country/territory has a score below 80, for example, on control of corruption, it means that the country has a relatively poor performance on corruption control, while countries/regions scoring above 80 have relatively better performance.

To compare the extent to which each governance indicator was examined in the scholarship in our corpus, we summed the number of documents that considered the topics associated with each governance category and calculated that category’s prominence. For example, topics related to performance management, human resource management (HRM), networks, local government and so on were grouped into the government effectiveness category, and the number of documents focused on all of these topics was used to calculate the prominence of governance effectiveness in the public administration literature. A number of our topics did not align with any World Bank Governance Indicator, and we categorised these topics as ‘unknown’.

3.1 Topics in the Corpus of Public Administration Journals

Table 2 presents the findings of our topic modelling analysis. It provides information on the topic rank (column 1), which was derived from the topic weight (column 5) and the number of articles associated with the topic (column 4). The topic names, developed by the expert panel, are given in column 2, and column 6 provides the top five terms for each topic generated by the topic modelling analysis. Figure 1 provides a visual depiction of the stock of topics in public administration between 1991 and 2019.

Figure 1 The topics in public administration

The highest ranked topic was health care, which accounted for 4.1% of all scholarships in our corpus of public administration journals. In contrast, training, which ranked as the fiftieth most common topic, accounted for just under 1% of the scholarship over the 1991–2019 period. The top five topics – health care, federal government, performance management, environmental regulation and HRM – accounted for just under a fifth (17.9%) of academic writing, while the bottom five – evaluation, risk management, conflict management, procedural justice and training – accounted for just over a twentieth (5.35%) of the scholarship. The top ten topics captured over one-third (34.6%) of all public administration scholarships. The bottom ten topics (ranks 41–50) accounted for just over a tenth (11.7%) of the scholarship.

The discipline of public administration is a design science that examines the administration and management of government and the implementation of policies to understand how public services can be delivered efficiently, effectively and equitably. Our results show, as might be expected, that the topics capture this focus, but with varying weights.

Some twenty topics examined questions of administration and management, accounting for 40.4% of all topics. Three of these topics were among the top ten topics: performance management (TW = 3.8%), HRM (TW = 3.0%) and networks (TW = 3.0%). Three other topics were associated with interorganisational management: networks (TW = 3.0%), collaboration (TW = 2.6%) and partnerships (TW = 2.2%). They were characterised by terms associated with management rather than structures, such as coordination, cooperation and projects. Three of the twenty topics were dedicated to government funding and its expenditure: public finance (TW = 2.0%), financing/budgeting (TW = 2.0%) and spending (TW = 1.6%). A number of topics were also associated with individual behaviour in bureaucracies, such as HRM (TW = 3.0%), senior civil servants (TW = 1.8%) and leadership (TW = 1.5%). Some individual management practices were also examined, including strategic planning (TW = 2.1%) and e-government (TW = 2.0%). This suggests that much of the scholarship in public administration is concerned with understanding the processes of management and administration within governments.

Looking first at the topics related to policy, our modelling identified eleven topics, of which ten focused on explicit policy areas and the remaining topic examined how policies are assessed, with a topic weight (TW) of 21.4%. The ten explicit policy topics were heath care (TW = 4.1%), environmental regulation (TW = 3.1%), science policy (TW = 2.7%), labour/employment policy (TW = 2.2%), development aid (TW = 2.0%), education (TW = 1.7%), economic policy (TW = 1.5%), inequality (TW = 1.5%), housing (TW = 1.3%) and family policy (TW = 1.3%). However, one limitation of topic modelling we conducted is that we were not able to directly correlate topics with published journal articles because the technique decomposes the articles into a topic matrix of words. We were thus not able to determine if these studies examined the efficiency, effectiveness and equity of policy or the processes of administration and management in these policy contexts. One example of this would be the extensive work of Kenneth J. Meier and Laurence J. O’Toole Jr. and colleagues, who developed a dataset of management and performance in the field of education policy. They have published on topics in managerial networking, personnel stability and management and their impact on organisational performance (Meier and O’Toole Reference Meier and O’Toole Jr.2003a, Reference Meier and O’Toole Jr.2003b), and their work has considered questions of the effectiveness and equity of public service delivery in school settings, thereby combining questions of management and policy consequences. One topic in the policy category, evaluation (TW = 1.2%), examined the assessment of policies. This suggests that just over a fifth of the scholarship in the corpus we examined was dedicated to questions of policy implementation.

Ten of the fifty topics were associated with the structures and processes of governance associated with structures, giving this area a TW of 18.7%. This included two of the top ten ranked topics, federal government (TW = 3.9%) and local government (TW = 2.6%), and the governance topics related to regional governance (TW = 1.6%), local decentralisation (TW = 1.5%) and multilevel governance (TW = 1.4%). Two of the structure topics also captured scholarship on the nonprofit sector – nonprofits (TW = 1.9%) and organisational autonomy (TW = 1.2%) – which were characterised by the topic terms voluntary, organisations and bodies. Two topics related to broad structural questions about the best arrangements for delivering public services were privatisation (TW = 2.4%) and contracting (TW = 2.2%), which were characterised by terms such as competition, firms and providers.

Under the broad category of accountability (TW = 5.6%), there were three topics: accountability (TW = 1.6%), which was characterised by the term responsiveness; citizen participation (TW = 2.7%); and user choice (TW = 1.3%). Based on topic names and terms, only two topics could be directly associated with more conceptual work in the discipline. These related to welfare regimes (TW = 2.5%) and the neoliberal paradigm (TW = 1.7%). As with our analysis of topics and their broad categories, the technique of topic modelling did not allow us to identify the specific articles associated with these topics and to fully understand if they are more conceptually or empirically orientated. However, at face value, the balance of the topics was associated with applied research.

In contrast, one topic examined the legal dimension of public administration, procedural justice (TW = 1.0%), while the topic of transparency (TW = 2.7%) drew on terms such as law, legal and rules and standards, suggesting that it had some association with the legal aspects of public administration. One topic had an explicit geographical title and focused on Europe (TW = 1.6%).

In summary, when we examined the stock of scholarship in public administration journals between 1991 and 2019, we found that (1) around 40% of all topics examined broad questions of administration and management, (2) just over a fifth topics related to policy, (3) just under a fifth related to questions of the structures and processes of governance and (4) a twentieth to accountability.

3.2 Topics in Individual Journals

We conducted a separate topic modelling analysis for each journal and identified thirty-five topics for each journal, rather than the fifty for the topics in the discipline because the corpora were smaller. To make inter-journal comparison easier, we focused on the top ten topics for each journal. These analyses had two purposes. First, we sought to examine the extent to which the topics identified in each journal’s sample of published articles were consistent with the journal’s stated aims (quotations below are taken from the website of each journal and the sources are included in the references section). Second, we compared each journal’s top ten topics with the topics in the whole corpus, seeking to determine within the stock of topics in all journals what were common across the stock of topics within journals and to what extent terms were shared in the common topics. (See the topic labels in Table 2 and Figure 1.)

Administration & Society seeks to further the ‘understanding of public and human service organizations, their administrative processes, and their effect on society’. Many of its top ten topics (see Table 3), including health care, education, welfare, citizen participation and gender and diversity, correspond with the journal’s main aim. The journal is particularly interested in ‘(1) studies that analyze the effects of the introduction of administrative strategies, programs, change interventions, and training; and (2) studies of intergroup, interorganizational, and organization-environment relationships and policy processes’. This specific focus is reflected in several of the identified topics, such as civil servant, nonprofits, new public management (NPM), HRM and governance outcomes.

Most of the journal’s top ten topics are consistent with the whole corpus. Two exceptions are NPM and governance outcomes, suggesting that the journal pays special attention to the impact of NPM reforms and governance outcomes. A comparison of the specific terms in the topics shared between the journal and the whole corpus suggests that the focuses are different. For example, in the education topic, the representative terms in the whole corpus are schools, universities and students, whereas those in A&S are federal programmes, schools and goals.

American Review of Public Administration focuses on ‘public administration broadly defined, publishing scholarship on all aspects of the field, including such areas as organization and management studies, program and performance evaluation, and budgeting and financial management, network governance, public involvement, and public service motivation’. The top ten topics are mostly consistent with the journal’s scope, especially networks, partnerships, HRM, public service motivation (PSM), strategic management and representative bureaucracy (see Table 4). A comparison of the specific terms of the topics shared by the journal and the whole corpus reveals some nuances. For example, for the strategic planning topic, representative terms in the whole corpus are rural, tools and spatial – terms that seem to suggest a relationship with urban planning, whereas in ARPA the terms include values, conflict and formal – terms that are more central to strategic planning in public administration.

Environment Planning C: Politics and Space seeks to ‘advance debates on the spatialization of politics and the politicization of spatial relations’. In other words, it focuses on the processes that make spatial and environmental issues politically relevant or contested. Among the top ten topics (see Table 5), climate change, energy, innovation and regional governance are closely related to the journal’s aims. The fact that the journal has four unique topics, including climate change, energy, innovation and entrepreneurship, indicates that it is a specialised journal focusing on these topics. However, the other topics, such as finance and budgeting, health care, public finance, privatisation and partnerships, seem somewhat unrelated to the relationship between political and spatial issues and may reflect the journal’s prior title, which included the subtitle Government and Policy rather than Politics and Space.

If we compare the terms in the topics shared between the whole corpus and this journal, we can identify some nuances. For example, in the public finance topic, while both corpora focus on taxes, the whole corpus focuses more on pensions and revenue, whereas EPC focuses more on incentives, rates and taxation – terms related to different ways to promote business development.

Governance provides a forum for ‘the theoretical and practical discussion of executive politics, public policy, administration, and the organization of the state’. Several of the top ten topics, such as delegation, corruption, transparency, supranational governance and multilevel governance, speak perfectly to the journal’s focus on executive politics and the organisation of the state (see Table 6). Topics related to finance and budgeting, service provision, economic policy and gender and diversity correspond to the journal’s interest in public policy and administration. The remaining topic seems to be related to organisational adaptation in China, demonstrating the journal’s international and comparative approach.

Comparisons of the terms in the topics shared by the whole corpus and this journal reveal some differences. For example, for the economic policy topic, the whole corpus focuses on investment, economies and scale, whereas Governance focuses on crisis, the IMF and water – terms that seem to reflect how different governing bodies like the IMF respond to crisis, reflecting the journal’s narrow focus within economic policy.

The International Public Management Journal features research on ‘public management and government reform, comparative public administration, organizational theory, and organizational behavior’. Several of the topics identified in the analysis, such as HRM, public finance, networks, contracting and red tape, are related to broad public management and government reform issues (see Table 7). There are also articles examining turnover, PSM and discretion, which are part of the organisational behaviour literature. The journal seems to focus on public management issues, such as coproduction, in the health and education fields.

Comparisons of the terms used to denote topics in the whole corpus versus this journal reveal nuanced differences. For example, for the contracting topic, the terms used in the whole sample (competition, cost, providers) primarily relate to the structural aspects and dynamics of the contracting process, whereas the terms related to this topic in the journal’s corpus (strategic, perceived, fiscal, opportunism) focus on the strategic and behavioural factors that influence contracting decisions and outcomes.

The International Review of Administrative Sciences claims to be one of the oldest journals in public administration. It focuses on ‘comparative and international topics’ in public administration, especially reflections on ‘international comparisons, new techniques, and approaches, academic-practice dialogue, and the future of public administration’. The topics in the journal speak to a wide array of public administration issues, such as decision-making (discretion, innovation, accountability), HRM (PSM, gender and diversity), financial management (spending) and intergovernmental or cross-sectoral relations (multilevel governance, federal government, privatisation) (see Table 8). However, this list of topics does not show whether these articles speak to the journal’s aim to address comparative and international topics, partly because of the limitations of the topic modelling approach, as already noted. A content analysis of the articles is needed to determine whether the articles carry out comparative and international research.

The specific terms used to define the topics shared by the journal and the whole corpus show some nuanced differences in their focuses. For example, for the government spending topic, the whole corpus uses terms related to the broader aspects and principles of government financial management (expenditure, fund, size), whereas the terms in the journal’s corpus specifically focus on the local context and specific areas of government spending (municipal, cities, police).

The Journal of Public Administration Research and Theory seeks to publish ‘organizational, administrative, managerial, and policy-based research that improves our understanding of the public sector’. Of the ten topics identified in the analysis, red tape, street-level bureaucracy and local government are closely related to administrative research; other topics, such as networks, accountability, goal ambiguity, gender and diversity, e-government and contracting, are important topics in public organisations and management. As regulation may also be related to policy, the relationship between policy-based research and the top ten topics is generally clear (see Table 9).

Comparisons of the specific terms used in the journal and the whole corpus again reveal some nuanced differences. For example, in the whole corpus, the terms for the e-government topic (adoption, technology, media, communication) focus primarily on tools, infrastructure and the processes involved in implementing electronic government initiatives. However, the terms associated with the e-government topic in the journal’s corpus (citizen satisfaction, trust, web) focus on the outcomes and citizen-centric aspects of e-government.

Local Government Studies focuses on ‘local politics, policy, public administration and management and governance’ and serves as an important ‘forum for dialogue and exchange on local government’. Three unique topics emerged from the analysis, namely, city council, election and central–local relations, demonstrating the journal’s emphasis on local politics (see Table 10). The articles in this journal also examine local governance issues related to collaboration (networks), finance and budgeting (public finance, spending), accountability, transparency, leadership and e-government.

A comparison of the terms representing the shared topics in the whole corpus and the journal’s corpus reveals interesting differences. For example, the leadership topic in the corpus is represented by terms like skills, style and transformational, which are related to individual characteristics, behaviours, skills and leadership styles. In contrast, in the journal corpus, the topic uses terms like officers and chief executives, which highlight the specific leadership positions that are responsible for decision-making within an organisational structure.

Public Administration is interested in articles on ‘all facets of public administration, public policy, and public management’, especially those that deal with ‘major administrative challenges that generate theoretical advances and provide substantive insights’. The top ten topics reflect the broad scope of the journal, which covers issues related to governance and regulation (networks, regulations, coalition, performance management and contracting), public sector management (PSM, innovation and science policy) and social issues (gender and diversity and health care) (see Table 11).

A comparison of the terms used to represent the topics shared by the whole corpus and the journal’s corpus shows some differences. For example, for the performance management topic, both corpora use the terms performance, measurement and indicators, indicating that these terms are integral to research on the performance management process. However, the two corpora also have some distinct focuses. The terms associated with performance management in the whole corpus include learning and improvement, highlighting the developmental and growth-oriented aspects of performance management. In contrast, the journal corpus includes the terms expectation and success, suggesting a focus on the alignment of performance with expectations and the attainment of successful outcomes.

Public Administration and Development states that it focuses on ‘public administration at the local, regional, national and international levels where it is directed to managing development processes in low- and medium-income countries’ and gives special attention to ‘the management of all phases of public policy formulation and implementation’ in the public and NGO sectors. Several of the topics identified in the analysis, such as corruption, solid waste, water and environmental regulation, are important problems faced by low- and medium-income countries (see Table 12). The topics also reflect the critical role of think tanks, nonprofits, privatisation, decentralisation, finance and budgeting and benchmarking in development.

An examination of the terms representing the shared topics in the two corpora shows similarities and differences. For example, for the environmental regulation topic, the terms in the whole corpus (regulation, sustainability, water) encompass the broader concepts and principles of environmental regulation, sustainability and water management, whereas the terms in the journal’s corpus (enforcement, welfare, pollution, green) highlight specific aspects of enforcement, social well-being, pollution control and environmentally friendly practices within the general topic of environmental regulation.

Public Administration Review focuses on a wide range of topics and is ‘the only journal in public administration that serves both academics and practitioners interested in the public sector and public sector management’. Some of the top ten topics reflect a broad focus on institutional actors and their interactions, such as local government, nonprofits, networks and citizen participation, whereas others examine specific aspects of public sector management, such as HRM (PSM, gender and diversity, motivation, civil servant) and financial management (finance and budgeting). Interestingly, knowledge of science emerges as a unique topic for this journal, suggesting that it pays special attention to fundamental discussions of knowledge and science in the field of public administration (see Table 13).

Comparisons of the specific terms in the shared topics in the two corpora – the whole corpus versus the PAR articles – reveal similarities and differences in the topics. For example, for citizen participation, both corpora use such terms as citizen, participation, involvement and engagement, indicating that these terms collectively emphasise the importance of citizens’ active participation in decision-making. However, the whole corpus focuses on how citizen participation is connected to democracy and legitimacy, whereas PAR articles highlight the interrelations between citizen participation and trust.

Public Management Review focuses on promoting ‘the dissemination and discussion of such research about public management … across the world’ and values inter-disciplinary work. The top ten topics in the PMR articles are HRM, innovation, performance management, networks, accountability, e-government, health care, strategic planning, sustainability and red tape (see Table 14). These topics represent different aspects of managing and improving public sector organisations, such as people management (HRM), monitoring and enhancing performance (performance management) and ensuring responsibility and transparency (accountability). While there is considerable overlap between the top fifty topics in the whole corpus and the topics covered in this journal, two topics stand out as distinct in the journal corpus: innovation and sustainability. This indicates that the journal has a unique focus on fostering creativity and progress (innovation) in public administration and the responsible use of resources, environmental stewardship and social responsibility (sustainability).

We find subtle differences in the specific terms that characterise the shared topics in the two corpora. For example, in the whole corpus, terms for the accountability topic (expectations, responsiveness, agents) highlight the expectations placed on individuals or entities, their ability to respond to those expectations and the entities’ accountability for their actions. In contrast, in the PMR corpus, the terms for the accountability topic (regulatory, transparency, charities) emphasise the mechanisms and practices related to ensuring accountability, including regulatory frameworks, transparency in processes and the specific accountability context of charitable organisations.

The aims and scope of Policy & Politics highlight research that contributes to ‘existing comparative policy literature, or to policy studies, and political science more generally via the use of comparative methods’. The articles in the journal are consistent with this aim. For example, health care, urban governance, environmental regulation, inequality and welfare regimes are specific, important policy topics (see Table 15). A unique topic of the journal – policy process – reflects its focus on ‘the public policy spectrum and all levels of the policy process’. Other topics in this corpus, such as networks, privatisation, partnerships and gender and diversity, are important issues in public administration and political science more generally.

There are some distinct differences between the terms representing the topics shared by the journal and the whole corpus. For example, for the urban governance topic, while both corpora focus on regeneration in an urban context, the topic in the whole corpus is represented by terms such as corruption, pay and collective, highlighting governance-related issues and practices, such as corruption prevention, fair compensation and collective decision-making, whereas the terms for the same topic in the journal’s corpus (areas, diversity, communities) reflect the specific focus on neighbourhoods, the diversity of populations and the importance of community engagement in urban governance.

Public Performance and Management Review ‘addresses a broad array of influential factors on the performance of public and nonprofit organizations’. The top ten topics are consistent with the stated themes of the journal, including improving budget strategies (finance and budgeting, performance budgeting), managing human resources (HRM, autonomy), building partnerships (partnerships, networks, contracting), facilitating citizen participation (trust, local government) and applying new technologies (innovation) (see Table 16).

A comparison of the common topics shows how the specific terms differ in the two corpora. The terms representing the local government topic in the whole corpus (politicians, elected, party) focus on the individuals, processes and organisations related to governance and political representation in local government, whereas the terms in the journal’s corpus (citizen, police, force) centre on community engagement, security and the roles of law enforcement agencies.

Public Money & Management covers finance, policy and management issues in public services. Among its top ten topics, accounting, finance and budgeting and cost benefits speak to the journal’s focus on public finance (see Table 17), reflecting its origins in UK public finance, accounting and auditing. Other topics, such as health care, partnerships, performance management, innovation, multilevel governance and privatisation highlight the journal’s attention to broader public management issues at different levels of government and across different sectors. Comparisons of the specific terms used in the whole corpus and the journal’s corpus indicate that they have distinct focuses. For example, in the multilevel governance topic, both corpora have Wales and devolution, which shows that the devolution from the United Kingdom to the Welsh government is an important political issue in the area. However, the terms associated with the topic in the whole corpora include integration and governing, indicating an emphasis on the concepts and processes involved in coordinating governance across different levels of authority in multilevel governance systems, while the journal corpus has terms like Scotland and Ireland, indicating its interest in specific regional contexts within such systems and the governance arrangements and processes of those regions.

Social Policy and Administration covers social policy issues not only in Europe but also in the United States, Canada, Australia and the Asia Pacific region. The topics clearly reflect the journal’s focus. For example, childcare, health care, migration, family policy, user choice, labour and employment policy and elderly care are important topics in social policy and administration (see Table 18). In addition, public finance, union and partnerships are topics that overlap with public administration. However, the international perspective is not apparent in the identified topics except the term postcommunist in the partnerships topic. Comparisons of specific terms that characterise the shared topics in the whole corpus versus the journal’s corpus reveal some interesting differences in their conceptual focus. For example, the terms characterising the family policy topic in the whole corpus (administrators, identity, family) focus on the institutions, concepts and dynamics related to policymaking and implementation in the context of families, whereas those in the journal’s corpus (parents, lone, workfare, mothers) pay more attention to various aspects of family structures, roles and support systems within family policy.

3.3 Comparing Journal Topics

We also examined which topics were common across the peer-reviewed journals in our corpus, or the stock of topics within all journals, and the variation in the journals’ specific focuses within these topics, or the stock of topics by journal. Table 19 lists the thirty-eight topics that occur in the topic list for the whole corpus and for at least one journal. Some topics were extensively studied in multiple academic journals, indicating that they are fundamental issues that public administration scholars consistently examine. For example, networks appeared in nine journals’ top ten topics. Gender and diversity and health care were among the top ten topics in seven journals. These recurring themes reflect a common consensus within the field. However, some topics, although ranked among the top topics in the whole corpus, were niche topics specific to particular academic journals. For example, family policy and labour and employment policy only appeared in the top ten topics list of SPA. Inequality and welfare regimes were only among the top ten topics for P&P. These specialised areas of research cater to the specific interests, aims, scopes and focuses of those journals and illustrate the variety of topics and academic publication outlets in the discipline.

To further examine how the widely studied topics were approached by different academic journals, we zoomed in on the topics that appeared in the top ten topic lists of at least five journals, that is, networks (in nine journals), gender and diversity (in seven journals), health care (in seven journals), partnerships (in six journals), HRM (in five journals), finance and budgeting (in five journals) and privatisation (in five journals). We examined the extent to which common terms are used among public administration journals by calculating the percentage of journals that shared the same terms in their topics. Specific terms in the topics are also listed by journals in Table 19.

Networks was the most widely studied topic. It appeared as one of the top ten topics in over 50% of the sample journals (in nine journals). The terms ‘network’ and ‘networks’ were among the top five terms characterising the topic in all of these journals (100%). Despite the common interest, these journals had specific focuses within the network topic. Some journals examined actors and governance structures in the networks, featuring terms like actors (PMR), agencies (P&P), structure (PMR), modes (PA), hierarchy (PA) and meta-governance (LGS). Some journals, especially IPMJ, paid more attention to resources in network management, as evidenced in the prominence of terms like capacity and resources. Other journals focused on network processes and dynamics – the interactions, collaborations and relationships within networks, represented by terms like coordination (ARPA, PMR), collaboration (PAR), interorganisational (JPART, PPMR), learning (PAR) and accountability (P&P). In addition, research on network was conducted in diverse contexts. Terms like disaster (ARPA), health (PPMR) and media (P&P) highlighted the specific domains or sectors where networks operate and the specific challenges and dynamics within those contexts.

Gender and diversity was the second most widely studied topic in our corpus of public administration journals, with seven journals featuring this topic as one of their top ten topics. Researchers examined the dynamics of gender equality and the representation of women and minority groups, as indicated by the terms gender (86%), women (86%) and representation (57%). A closer look at the specific terms in the journals’ topic definitions revealed the different lines of research and the nuanced difference in the journals’ focuses. For example, some studies examined how public administrators, especially elected officials (A&S, Governance), police officers (PA) and professionals (A&S) treat minorities (JPART), men (PAR) and public service users (IRAS) in different ways in various settings, such as employment (PAR), career development (IRAS) and public service delivery practices (P&P). These studies focused on how these differential treatments are related to differences (PAR), equality (P&P) and satisfaction (IRAS). While some journals, such as IRAS and PAR, focused on representation in hiring and promotion in the workplace, most journals, such as A&S, Governance, JPART, PA and P&P, examined how different minority groups are treated in service delivery.

Similar to the gender and diversity topic, seven public administration journals had health care as one of their top ten topics. These journals commonly used the terms health (71%) and care (86%), with particular attention paid to the United Kingdom’s National Health Service (NHS) (57%). Despite the common focus, these journals focused on different issues in health care. For example, journals like P&P and SPA focused on how to ensure the quality of care and used specific criteria to evaluate the effectiveness of care provided by medical practitioners, particularly in the realm of mental health. Some journals, such as PMM and EPC, paid more attention to health care reforms, as evidenced by the frequency of terms like devolution and decentralisation. The decentralisation reform in the NHS, part of which revolved around the establishment of trusts, has had a significant impact on the delivery of welfare and health care services. Other journals, such as PMR and A&S, focused on how managers design and implement strategies and policies, such as the involvement of health care professionals and patients, to enhance patient care and outcomes.

Partnerships featured as a topic in six journals, and the terms partnership (83%) and its plural form (100%) appeared frequently in the top five terms characterising the topic. By looking into the specific terms that represent the topic in different corpora, we found that a recurring theme in the journals was stakeholder collaborations, such as those among communities, businesses, members and parents, in diverse settings, such as regeneration. Another line of research examined procurement, bids and privatisation – the technical and financial aspects of partnerships. Compared with studies on other topics (e.g., networks), which were often quite diffuse, our analysis suggests that studies on partnerships were relatively focused.

Five journals covered HRM topics extensively, and the terms shared by the topic in these journals included job (100%), satisfaction (80%) and employee (60%), reflecting the topic’s overall focus on employee and job satisfaction. Zooming in on the specific terms representing the topic, we found that these journals consistently focused on employee attitudes, commitment, satisfaction and emotional labour in the workplace. There were also some studies of leadership, diversity and person–organisation fit, and how these factors shape job attitude and performance.

Finance and budgeting figured extensively in five public administration journals. Fiscal (80%) and budget (60%) were two common terms representing the topic. A detailed examination of the specific terms across journals showed that studies on finance and budgeting were clearly focused on either the broader management of public finances, such as revenue generation (tax, outsourcing, stress), expenditure control (expenditure, spending) and overall financial policy (decentralisation, intergovernmental, innovation, adoption), or the allocation of financial resources, such as creating, implementing and monitoring funds.

Privatisation was a topic included in the top ten topics of five public administration journals, with the term privatisation appearing in three journals’ topic terms. However, the specific terms used in different journals indicated the different lines of research. One line of research examined the broader economic impact of privatisation, including globalisation, liberalisation and efforts to enhance competitiveness. Another line of inquiry focused on regulatory frameworks and the stakeholders involved in the privatisation process, such as businesses, elites and recipients. The studies of privatisation were conducted in diverse sectors, such as telecommunication, transportation, agriculture, infrastructure and energy.

All seventeen journals published articles on the seven most popular topics in the inter-journal analysis. Policy and Politics included articles on these inter-journal topics most frequently, on five occasions (networks, gender and diversity, health care, partnerships, privatisation). Two journals published four of the topics: PAR (networks, gender and diversity, HRM, finance and budgeting) and PPMR (networks, partnerships, HRM, finance and budgeting). Four journals published on two of these topics: Governance (gender and diversity, finance and budgeting), IRAS (gender and diversity, privatisation), PAD (finance and budgeting, privatisation) and SPA (health care, partnerships). This suggests a wider diversity of published topics in these journals, seen by single topics in these journals such as federal government (IRAS), decentralisation (PAD) and family policy, labour and employment policy and user choice (SPA).

4.1 Changing Topic Weights

As the number of published articles varied significantly across years, we calculated the topic prominence by using all of the articles on a certain topic published in a year divided by the total number of articles published annually in the seventeen journals for the 1991–2019 period. This presents the flow of topics over time. Figure 2 presents the changes in the relative prominence of the top ten topics over the study period. On average, these topics were examined in approximately 3% of the articles every year, with peaks and lows for different topics.

Figure 2 Evolution of the top one to ten topics in public administration

Health care emerged as a highly topical subject during the late 1980s, with approximately 15% of all articles dedicated to health-related topics. This surge can be attributed to the widespread health care reforms implemented in the United Kingdom and other nations. However, in the subsequent decade, the prominence of health care experienced a noticeable decline, with minor fluctuations between 1995 and the early 2000s. Health care issues regained prominence in 2008, likely due to the implementation of the Obama Care reform in the United States. The prevalence of this issue gradually diminished after 2010.

Research on the federal government gained momentum in the 1990s and reached its zenith in 1995. Subsequently, there were fluctuations in the number of studies, culminating in another significant peak in 2008. This resurgence can be attributed to the profound impact of the Great Recession, which ignited fervent debates on the role of federal government in both economic development and governance.

Research on performance management increased significantly in the 1990s and peaked in 1995. After small fluctuations between 1995 and 2013, the relative number of studies on performance management have been growing at a rapid rate since 2014. This sharp growth can be largely attributed to the series of government reforms in performance management in different countries following the NPM movement. For example, in 2013, the US government promulgated the GPRA Modernization Act, which retains and amplifies some features of the Government Performance and Results Act of 1993 (GPRA 1993) while also addressing some of its weaknesses. The more recent growth in this topic could be accounted for by the advent of behavioural public administration, much of which focuses on the use of performance information in public organisations and by citizens.

The prominence of research on environmental regulations has been relatively stable, except for the sharp growth between 1990 and 1995, probably because of the dedicated policy attention to environmental issues, as evidenced by the Rio Earth Summit in 1992 and other international events, which triggered increased global awareness and action to address environmental issues and the development and implementation of various regulations and agreements. In addition, interest in environmental regulations peaked a second time around 2016, probably because of the Paris Agreement and other global environmental treaties.

Research on HRM has slowly gained prominence since the 1990s and gained momentum in 2002, probably because the Federal Employee Viewpoint Survey was first conducted in 2002 to assess the views and experiences of federal employees. This suggests that data availability is critical to scholarship.

Research on networks was very popular in the 1980s and declined sharply in the 1990s, after which the relative volume of studies on the topic has been stable. Interestingly, the pattern for health care and networks was similar: both experienced a noticeable decline after a period of prominence in the 1980s. The prominence of articles on citizen participation and transparency exhibited similar patterns: substantial growth in the 1990s, followed by minor fluctuations in later years.

Studies of science policy increased around 1993, and then quickly declined. The peak can likely be attributed to the development of the Information Superhighway in the United States, as evidenced by the significant expansion of Internet access, advancements in digital technologies and the rise of the World Wide Web. During this period, there was a surge of public interest in and policy discussions surrounding the development and use of the Information Superhighway. Interest in local government gained momentum in 2002, following a similar pattern to that of HRM.

Figure 3 shows the flow of the evolution of the topics ranked as the eleventh to twentieth most popular in the whole corpus. Interest in research on welfare regimes fluctuated every two to three years and then gained momentum in 2005, probably because of the welfare regimes introduced in many countries in 2002 and 2003, such as the social assistance programme and urban social insurance reforms in China. Studies on culture and values were very prominent at the beginning of our study period but interest sharply declined in the early 1990s, followed by downward fluctuations, suggesting that interest in this area of research is shrinking.

Figure 3 Evolution of the top eleven to twenty topics in public administration

Research on privatisation, collaboration, partnerships and contracting has remained relatively stable, with small fluctuations, probably because these topics are fragmented. Interest in representative bureaucracy has shown a slow, steady increase, suggesting that this topic is becoming more popular in the field as questions about diversity, gender and race figure prominently in the public discourses of many societies and are being reflected in scholarship.

Research on labour and employment policy experienced a dramatic decline in the early 1990s, followed by a relatively stable development. Two reasons may have contributed to this trend. First, certain social, policy and welfare changes may have promoted interest in labour and employment policy in public administration in the late 1980s to early 1990s. Second, scholarship on labour and employment policies may have moved from public administration to other fields, such as economics, leading to the decline in the number of studies published in public administration journals.

The number of studies on strategic planning has remained relatively stable, with small fluctuations. Interestingly, small peaks occurred approximately every five years, suggesting that it often takes time for strategic problems and issues to emerge and for strategies to be developed and implemented. Interest in e-government has also been relatively stable, but with a noticeable peak in 1997. Certain significant events and milestones related to e-government, such as the use of World Wide Web and the Government Paperwork Elimination Act in the United States (1998), may have laid the foundation for the digital transformation of government services and the adoption of e-government practices.

Figure 4 shows the evolution of interest in the twenty-first to thirtieth most popular topics in the whole corpus. Research on gender and diversity experienced a moderate decline in the early 1990s and remained low for about five years, bouncing back in 1997. The number of studies on the topic has remained relatively stable since the 2000s, with minor fluctuations. The topics of public finance, finance/budgeting and development aid exhibited largely similar patterns: small peaks in the 2000–2005 period, followed by stability. Research on nonprofits had a large peak in 1993, followed by small peaks in 2000 and 2015.

Figure 4 Evolution of the top twenty-one to thirty topics in public administration

Interest in the topic of senior civil servants was low throughout the study period, except for peaks in 1990 and 2010. The two peaks may have been driven by reform efforts made in those years to enhance the professionalism, effectiveness and accountability of senior civil servants.

Emergency management had generally low prominence but has experienced sharp growth since 2013. The combination of increased disaster events, the impact of climate change, technological advancements, policy changes and a growing recognition of the interdisciplinary nature of emergency management may have contributed to the rise in the relative number of studies on emergency management after 2013.

Research using a neoliberal paradigm was very popular in the late 1980s and early 1990s. As an important theoretical foundation for NPM, the neoliberal paradigm was heatedly discussed at that time. However, as interest in NPM waned, studies of the neoliberal paradigm declined significantly and have remained low since the 2000s, with small peaks occurring every five years. However, work has continued in a number of topics related to aspects of NPM, such as performance management. Research on PSM and red tape was relatively stable from the 1990s to 2017 but has increased significantly since then. The growing number of studies on PSM since 2017 indicates the reviving interest in the field. The relative volume of studies on security remained stable across the study period.

Figure 5 shows the evolution of the thirty-first to fortieth most popular topics. The relative prominence of research in education, urban governance, local decentralisation and leadership exhibited similar patterns: stable with minor fluctuations. The relative prominence of studies on regional governance and urban governance peaked at nearly the same time – around 2000. Research on accountability and inequality showed similar patterns, with some peaks in the early 2000s, and substantial growth after 2015. The surge in research on accountability and inequality may be driven by global governance reforms, international development agendas such as the UN Sustainable Development Goals and high-profile scandals.

Figure 5 Evolution of the top thirty-one to forty topics in public administration

Research on spending peaked in the 1990s, followed by a sharp decline in the late 1990s and then a stable pattern. Interest at the peak was probably associated with the prevalence of NPM, a management philosophy and set of principles that aim to improve the efficiency, effectiveness and accountability of public sector organisations, including their use of financial resources. Research on economic policy has been decreasing overall in the public administration field, probably because the topic has become more salient in other fields, such as economics. Research on Europe was relatively stable throughout the study period, with a peak around 2007, probably because of the several key events, such as the Treaty of Lisbon, EU enlargement, the 2008 financial crisis and ongoing debates on European identity.

In Figure 6, we show the evolution of research in the ten least popular topics. Research on multilevel governance, family policy, risk management and procedural justice has been relatively stable with minor fluctuations. The prominence of research in housing, organisational autonomy, evaluation and conflict management exhibited similar patterns: a sharp decline in the early 1990s, followed by small fluctuations. The large volume of studies on organisational autonomy and evaluation in the early 1990s was probably related to the NPM reforms. Studies on training remained rare throughout the study period except for a notable peak around 2004.

Figure 6 Evolution of the top forty-one to fifty topics in public administration

4.2 Keyness of Early and Late Period Topics

We also examined changes in the topics of public administration by performing a keyness analysis – the aboutness or the most common themes in a text corpus as represented by keywords and their LLR and p-value – of the early and late periods using a computational social science approach. In our corpus linguistics analysis, we first calculated the LLR of keywords to identify words with a p-level <.001 and a critical value of 15.13 (Rayson Reference Rayson2008). We used WordSmith software to examine the keywords with an unusually high log likelihood of appearing in articles in the early (1991–1995) and late periods (2015–2019) in the different corpora.

As shown in Table 20, during the early period, the corpus focused on issues, experiences and lessons in the context of the United Kingdom, particularly in the 1980s. In the late period (2015–2019), discussions revolved around issues of governance, collaboration, citizens, networks, employees, municipalities, actors and practitioners. These topics collectively highlight key concepts and themes in public administration studies, including collaborative governance, citizen participation, local governance and the practical implications of research for real-world challenges. In the next section, we present the findings of the keyness analyses of individual journals.

4.3 Keyness of Early and Late Period Topics in Select Public Administration Journals

We were not able to conduct keyness analyses for all of the journals because eleven of the seventeen journals in our corpus did not begin publication until after 1991. We therefore examined EPC, PA, PAD, PAR, P&P and SPA. Table 21 presents the results of the journal-level keyness analysis of all articles published by EPC. In the early period, the corpus did not have a clear focus, whereas in the late period, key themes emerged, including governance, climate, water, environmental, networks and actors. These keywords reflect the journal’s focus on politics and space, highlighting the interdisciplinary nature of environmental governance, which involves multiple stakeholders, policy domains and complex environmental challenges, as well as the importance of collaboration, policy coordination and the active participation of key actors to address environmental and climate issues.

Our keyness analysis of PA showed that in the early period, the journal discussed issues relating to expenditure, Britain and the NHS, reflecting the massive health care reform implemented in the United Kingdom and issues related to government expenditures (Table 22). In the late period, the journal focused on coordination, performance and citizens, demonstrating the importance of coordination among diverse actors, performance assessment for effective governance and the role of citizens as key stakeholders in shaping public policies and administration. The keyness analysis captured the journal’s focus in the late period on ‘facets of public administration, public policy, and public management’.

The keyness analysis of the articles published by PAD in the early period showed that the corpus was characterised by terms like planning, environmental, projects, agricultural, training and privatisation (Table 23). These terms collectively highlight various dimensions in the field of development, including the importance of planning for sustainable development, the integration of environmental concerns, the implementation of development projects, the significance of agricultural development, the role of training in capacity building and the potential impacts of privatisation on development outcomes. In the late period, the journal became more focused, and the key terms included terms like police, policing and governance, illustrating the journal’s considerable attention to policing and governance issues. These topics somewhat contrast with the aims of the journal, which are to examine ‘the management of all phases of public policy formulation and implementation’.

Table 24 presents the results of the keyness analysis of the articles published in PAR. In the early period, administrators stood out as a keyword in the corpus. However, in the late period, the journal focused on issues related to performance, citizens, police, actors, evidence and collaboration, which collectively show the journal’s focus on a broad range of issues in public administration, such as performance measurement and management, the role of citizens, the management of law enforcement agencies, the use of evidence in decision-making and collaborative approaches to governance and service delivery. The keyness analysis reflected the journal’s current broad aims and scope.

Our keyness analysis of P&P revealed a strong focus on policy issues, in particular those related to health and housing, between 1991 and 1995 (Table 25), which shifted in the late period to issues of media, governance and public, highlighting the journal’s attention to the role of the media in shaping policy discourse and public opinion, the significance of governance in policy processes and the centrality of the public as both recipients and active participants in public services and policies. These keyness terms in the late period clearly captured the journal’s focus on policy studies but did not tease out its comparative research and methods aims.

The final journal that published throughout our study period was SPA and our analysis showed a reduction in the number of keyness terms from three to one – the only other journal to show this pattern was PAD (Table 26). During the early period, the corpus was characterised by the terms medical, police and NHS. These terms collectively demonstrate the importance of social policy in addressing health care and public service-related concerns. In the late period, one keyword emerged from the corpus – countries – suggesting that the journal paid considerable attention to cross-country comparison in studies in that period, reflecting the journal’s aims to examine the social policies of different countries.

5.1 World Bank Governance Indicator Categories and Topics

The proportion of the corpus focused on each of the six World Bank Governance Indicator categories broadly reflected the overall pattern of topics in the corpus, as shown in Table 27. The rank order of the six indicators (Table 28) was government effectiveness, voice and accountability, regulatory quality, rule of law, political stability and control of corruption. However, there were variations in topic weights for countries with low and high governance scores. In our analysis, we highlighted differences in topic weights equal to or greater than 1% between low-and high-scoring countries and within each of the indicators. Here, we discuss differences in the top five topic terms. (A detailed analysis of the topics within each indicator is presented in Appendix tables A1–A6.)

In countries/territories scoring low in control of corruption, government effectiveness accounted for nearly 50% of research by topic weight (49.9%), which was 1.8% higher than in high-scoring countries. Among the top five topics in each indicator, scholars in lower-scoring countries prioritised research on local government (1.5% higher TW than in high-scoring countries), performance management (0.5% higher TW than in high-scoring countries), networks (0.3% higher TW than in high-scoring countries) and federal government (0.2% higher TW than in high-scoring countries). Government finances (finance/budgeting) were only a top five topic in low-scoring countries, and people management (HRM) was only a top five topic in high-scoring countries.

Differences of over 1% in topic weights were also observed for the voice and accountability and political stability indicators. In the political stability indicator category, scholars in high-scoring countries focused on three topics: security, risk management and conflict management. In contrast, scholars in low-scoring countries placed more emphasis on accountability (0.9% higher TW), representative bureaucracy (0.8% higher TW), transparency (0.7% higher TW) and citizen participation (0.6% higher TW), suggesting that scholars working in countries that score low on government corruption examine voice and accountability and the effectiveness of government to a greater extent than those in high-scoring countries, with perhaps a focus on improving the quality of government.

In the high-scoring subsample, topic weights in the government effectiveness indicator were higher for government effectiveness (1.7% higher TW) and for rule of law (1.1% higher TW) than in the low-scoring countries. There was a stronger focus on voice and accountability (1.5% higher TW) in the low-scoring subsample than in the high-scoring subsample. While scholars in high-scoring countries dedicated more attention to government effectiveness, the highest topic weights in the low-scoring subsample were associated with federal government (4.7% vs. 3.9%) and local government (3.7% versus 2.5%), suggesting a focus on structures. In contrast, scholars in high-scoring countries placed more emphasis on performance management (3.9% versus 2.8%) and included HRM in the top five topics, implying a stronger interest in questions of management. There were no substantial differences between the two subsamples in the prominence of different governance dimensions associated with regulatory quality, political stability and control of corruption.

The political stability indicator showed the least variation between low- and high-scoring countries, with negligible differences across the topic weights, with one exception. The gap in the topic weights between low- and high-scoring countries was largest for the government effectiveness indicator, at 4.6%. An emphasis on structures was again seen in the low-scoring subsample: federal government (TW of 4.1% versus 3.5%) and local government. It was not even a top five topic for high-scoring countries. In addition, two management topics were present in the top five topics in low-scoring countries – HRM and performance management – but they had lower topic weights than in high-scoring countries. The networks topic was more prominent in high-scoring countries (TW of 3.5% versus 2.65%). These findings again point towards interest in research questions on government structure in low-scoring countries.

When we compared the distribution of topics related to the regulatory quality indicator, we found that government effectiveness (1.3% higher TW) and voice and accountability (1.6% higher TW) had higher topic weights in high-scoring countries than in low-scoring countries, and the opposite pattern for the rule of law (1.9% higher TW). Within the government effectiveness indicator, structures was again more common in low-scoring countries: federal government had a 0.7% higher topic weight in low-scoring countries than in high-scoring countries, and there was a 0.4% higher topic weight for both local government and networks. Management was also featured in low-scoring countries, with performance management having a 0.1% higher topic weight than in high-scoring countries. High-scoring countries also conducted more research on the topic of HRM.

For the rule of law, two governance indicators presented differences in topic weights of over 1%: government effectiveness was 1% higher in high-scoring countries and voice and accountability was 1.5% higher in low-scoring countries. Government effectiveness was the most prominent governance dimension discussed in both subsamples, followed by voice and accountability, regulatory quality, political stability and rule of law. Within government effectiveness, the emphasis on structures was again seen in low-scoring countries: topic weights were higher for local government (1.4%), federal government (0.8%) and networks (0.4%). Topic weights were higher for performance measurement in high-scoring countries (0.4%). PSM and red tape was a top five topic in low-scoring countries and HRM was one in high-scoring countries. For the voice and accountability indicator, low-scoring countries conducted more research than high-scoring countries on accountability (1.1% higher TW), representative bureaucracy (1% higher TW) and citizen participation (0.1% higher TW) but less on transparency (0.1% higher TW). User choice was among the top five topics in low-scoring countries only, whereas gender and diversity was only included in the top five topics in high-scoring countries.

For the final World Bank Governance Indicator, voice and accountability, differences of over 1% in topic weights were seen for government effectiveness (1.4% higher TW in low-scoring countries) and rule of law (1.1% higher TW in high-scoring countries). The focus on government structures was again higher in low-scoring countries: local government (1.1% higher TW) and networks (0.3% higher TW). However, scholars in high-scoring countries examined federal government more frequently (0.2% higher TW). Scholars in low-scoring countries focused on performance measurement slightly more than those in high-scoring countries (0.2% higher TW). PSM and red tape was among the top five topics by weight in low-scoring countries and HRM was a top five topic in high-scoring countries.

The variations in topic weights given to World Bank Governance Indicators across the groups were not large. However, there were important differences in the research agendas of countries/regions with low and high governance scores. In countries/regions that score relatively low on these indicators, there is often an emphasis on research that examines the structures of government, whereas in countries/regions with high governance scores, scholarship is geared more towards questions of management.