Trends in research approaches and gender in plant ecology dissertations over four decades

Dissertations are a foundational scientific product; they are the formative product that early-career scientists create and share original knowledge. Methodological approaches used in dissertations vary depending on the research field. In plant ecology, these approaches include observations, experiments (field or controlled-environment), literature reviews, theoretical approaches, or analyses of existing data (including ‘big data’). This dataset was created to examine how the emphasis on each of these categories has changed over time, and whether male and female authors differ in the methods employed. The dissertations used for this study were gathered from the Proquest Dissertations and Theses Global (https://www.proquest.com/pqdtglobal) database.

Methods
The Proquest Dissertations & Theses Global Database (https://www.proquest.com/pqdtglobal) was used to find relevant dissertations. The website was accessed on January 4, 2022, and the following search string was used:

((SU(plant? OR vegetation OR tree OR leaf OR botan* OR flora* OR seedling* OR grass*) OR TI(plant? OR vegetation OR tree OR leaf OR botan* OR flora* OR seedling* OR grass*)) AND (SU(ecology OR ecolog* OR ecosystem? OR communit* OR conservation OR diversity OR biodiversity OR range OR trait?) OR TI(ecology OR ecolog* OR ecosystem? OR communit* OR conservation OR diversity OR biodiversity OR range OR trait?)) AND LA(en OR eng OR english)) NOT ("water reclamation" plant? OR "water treatment" plant? OR econom* OR bioengineer* OR biotechnolog* OR "bacterial flora")

The search results were further filtered as follows:

Manuscript type: Doctoral dissertations; Language: English

Subject: NOT (plant pathology AND genetics AND agronomy AND microbiology AND soil sciences AND zoology AND plant propagation AND molecular biology AND business community AND livestock AND wildlife management AND animal behavior AND physical geography AND cartography AND fish production AND behavioral sciences AND civil engineering AND enzymes)

This search returned 5423 results. These were then manually screened for metadata completeness (e.g., author name, year of publication) and availability of abstract. Dissertations with incomplete information, particularly missing abstract or year of publication, or those which were not in English, were removed. This left us with 3832 studies. These were then screened for relevance, based on whether the study focused on a topic in plant ecology. This was done by reading titles and abstracts. Relevant studies were defined as those that focused on one or more plant species or communities, and that studied the interactions of those plant species/communities with other organisms or with the environment. Only studies on embryophytes (i.e., bryophytes and vascular plants) were included. Other taxonomic groups that have been traditionally included under the term “plants”, such as algae and fungi, were excluded. After screening for relevance, 2670 dissertations remained.  

Methodological Classification

We initially selected 20% of the relevant dissertations at random for classification by category. After this initial classification, an additional 5% were added; as the proportions among categories were stable at that point, we felt that this was a sufficient sample of the approaches taken for the total population of studies for each decade. In total, 670 samples were classified.

Classification was carried out primarily by reading the abstracts. However, if the abstract did not contain sufficient information for unambiguous classification, the full text was then used, if available. If a thesis had both insufficient information in the abstract and lack of full text availability, it was removed and replaced by another randomly chosen dissertation from the same decade.

The classified dissertations were divided into 10-year time periods based on year of publication, for quantifying temporal trends. Note that studies from 2020-2021 were placed in the 2010s time-period. Studies from before 1980 were grouped into a single period, due to the small number of such studies.

Gender Classification

The gender of each thesis author was determined using genderize.io (https://genderize.io/), an application programing interface (API) that uses social media records to predict a person’s gender from their given name. Along with a gender classification, the program also provides the probability of the name belonging to that gender, and the number of past records of that name in the API’s database. We set a probability cut-off of 0.8 and a past record cut-off of 50, to reduce the chances of false classifications (names that fell below either of these cutoffs were considered undermined). For authors whose gender could not be determined by their first names, we re-ran the algorithm on their middle names. if available. If middle names were not available, or if gender could not be determined from the middle name, we looked up online profiles (e.g., institutional profile, lab website, etc.) of the author, and attempted to determine gender from their photographs. Any remaining undetermined records were removed from the gender analysis. We were able to determine the gender of 2392 authors, out of a total of 2760 (ca. 87% of the dataset).

A Chi-squared goodness of fit tests was used to determine whether these gender ratios in each decade significantly differed from a 1:1 ratio (equal male and female representation). We also considered whether different methodologies were more or less likely to be used by either gender, using the dissertations that had been classified by methodology. This was done by comparing the gender ratio of authors in each methodology category to the overall gender ratio, using Chi-squared goodness of fit tests.