Rhiannon @ WPREU
Rhiannon shares her experiences interning for WPREU in both 2017 and 2018.
Walking into my placement with University of Sheffield's Widening Participation Research and Evaluation Unit (WPREU), the ethnicity attainment gap in UK universities was an issue that I really knew nothing about. I had heard bits here and there about some people arguing that it existed, others arguing that it was simply a matter of socioeconomic and other traits in the students’ backgrounds only making it look like an ethnicity-based issue; a matter of misinterpreted correlation. In all honesty, it really wasn’t a subject that I had really thought about all that much before - after learning more about it, however, it became something which I am interested in researching in the future of my academic career.
Some of the most commonly heard arguments against the presence of an ethnicity-based attainment gap are that black and minority ethnic (BME) students simply come in with lower entry grades or differing entry qualification types and their lower attainment rates of a 2:1 / 1:1 when leaving university are simply a reflection of that, or that they largely come from lower socioeconomic backgrounds and are therefore being affected by their background of relative deprivation; nothing to do with their ethnicity as an isolated factor. Other variables used to explain away the gap includes age, gender and subjects studied.
Previous work on the project had already confirmed that when previous attainment qualification types (A-level, BTEC, etc) are controlled for an ethnicity attainment gap remained with, as expected, white students always attaining higher grades. Much the same again in the case of entry grades, when controlled for, the attainment rate of a 2:1 / 1:1 degree classification fell far below the attainment rate for white students.
I learnt how to be selective with the data I chose to present and only include the data visualisations that directly explained the key findings of the report"
As important as this previous work on the project was, there were still arguments from ethnicity attainment gap deniers left over which had not been addressed; as such, myself and another student working on the project chose to look at gender, mature status, Polar3 (a measure of higher education participation rates in the students’ social background) and IMD ranks (a measure of socioeconomic deprivation), with myself investigating the latter two.
Sadly, though not surprisingly, none of these factors could sufficiently explain away the existence of an ethnicity-based attainment gap at The University of Sheffield; when these four factors were controlled for separately, BME students overwhelmingly still attained a 2:1 / 1:1 degree at lower rates than white students did. This spanned across all faculties in the university too; some faculties being worse than others, but all experiencing an ethnicity attainment gap.
Some of the findings of the project have also reinforced the notion that the way in which the ethnicity attainment gap is studied needs to be addressed within itself; namely, the fact that students of different BME subcategories attained a 2:1 / 1:1 at differing rates from each other and as such cannot be looked at as one homogenous group. This finding is as of yet unexplained although I do hope that it will be factored into more effective research designs, including the upcoming regression analysis work, for BME students’ issues to be addressed faster and more effectively.
Weeks 1 and 2
My 2018 placement organised via the PlaceME@SMI scheme is based in the Widening Participation Research and Evaluation Unit (WPREU) at The University of Sheffield.
This is my second time working with WPREU. The first time was in 2017 when I was utilising SPSS and Excel to create crosstabs and basic visualisations using a dataset of 16k+ student records to illuminate trends in degree classification attainment gaps amongst UK-domiciled undergraduate students who graduated between 2010/11 and 2014/15. In particular, I was investigating how attainment gaps correlating with socioeconomic background indicators interacted with the Black and Minority Ethnic (BME) attainment gaps.
The aim of that first placement was to investigate whether or not the apparent BME attainment gap, measured by final degree classifications, could be explained by any alternative attainment gaps. For example, if students from more deprived socioeconomic backgrounds were awarded a 1st or 2:1 at lower rates than students from more privileged backgrounds and the BME student body was comprised of a higher proportion of socioeconomically deprived students, the higher proportion of socioeconomically deprived students within BME student groups could be creating the appearance of a BME attainment gap which was actually caused by socioeconomic deprivation.
The finding was that the BME attainment gap could not be explained away by any alternative gaps; however, only one alternative attainment gap (eg socio-economic, gender OR age) was controlled for at any time, leaving the argument open that the BME attainment gap could still be explained by controlling for a combination of these other factors simultaneously.
As the weeks progress, I am steadily getting to the point of understanding all of the major (and most of the minor) aspects of binomial logistic regression"
Due to this limitation, my second placement with WPREU is based on improving upon my previous research by using more advanced statistical analysis techniques to assess whether or not controlling for a combination of confounding variables (demographics, socioeconomic indicators etc.) actually could explain away the so-called BME attainment gap. During the 3 - 4 weeks ahead, I will be improving upon my (basic) logistic regression knowledge to apply this technique to the same dataset I used previously - this means reading a ton of academic articles and guides over the next few weeks and putting in a lot of practice to ensure I understand every aspect of the method before using it.
By the time the first two weeks of the 6-week placement are over I am still stuck headfirst in a digital pile of articles, resisting the urge to face-palm at how awfully half of them are written - such is life for any researcher I suppose.
Weeks 3 and 4
As the weeks progress, I am steadily getting to the point of understanding all of the major (and most of the minor) aspects of binomial logistic regression. The word 'multicollinearity' used to make me cringe - now I understand the issue so clearly that testing for it and understanding the results are an easy breeze. However easy testing for it may be, though, deciding what to do with less than ideal results brings me to the most consistent barrier of all that I have found when conducting quantitative research - dealing with complications in the dataset which mean you cannot run your analyses the way you wanted to.
Finding that two of the main variables I would ideally control for are too-highly correlated to be included in the same regression model with one another means that I have to re-assess the situation before I can continue. There are multiple options at this junction – but do I transform the correlated variables (which takes up valuable time - deciding how to transform them, re-testing how they fit into the models and potentially having to transform them once again if they are still too highly correlated) or do I just save my time and remove one of them? In the latter case, how do I decide which variable to remove when both of them are theoretically important?
Articles, textbooks and guides are all fantastic resources when learning how to conduct quantitative research but when it comes down to roadblocks such as this in which there are no clear 'correct' answers, they can never make those difficult decisions for you.
By the end of week 4 of my placement with WPREU, I feel confident in my ability to use the logistic regression method in SPSS but am still unsure as to whether transforming or removing one of the too-highly correlated variables would be the best way to go. With only two weeks left to conduct the regression, write up the results and prepare a presentation based on them I am definitely feeling eager to start analysing; although not only because I am feeling stressed for time, but also because I am genuinely interested in seeing the results of my work.
I have become more proficient at using SPSS, become familiar with advanced statistical analysis methods and independently conducted research using those advanced methods"
Weeks 5 and 6
By the middle of week 5, I had settled on removing one of the problematic variables - a less than perfect solution, but it freed my hands to (finally) follow through on my analysis. With only a week and a half left, I was definitely pressed for time - although I was already familiar by this point with interpreting SPSS output and I already had a fairly well-developed idea of how to write up the report, a week and a half still does not seem like much time to run, interpret and write-up the results of 3 different logistic regression models. I managed to surprise myself by doing exactly this with a level of proficiency which I am proud of, however, leaving WPREU with a well-detailed report which will (I very much hope) aid them in their aims to help students from disadvantaged backgrounds have equal access to and opportunities during their university careers.
Having tackled a whirlwind of researching analysis methods, research planning, research re-planning once I ran into roadblocks, analysis, interpretation etc one day at a time, it is surprising how quickly the 6-week placement passed by. It is only at the end of the 6 weeks that I realise how much experience I have gained; I have become more proficient at using SPSS, become familiar with advanced statistical analysis methods and independently conducted research using those advanced methods knowing that the findings genuinely do matter and (hopefully) may actually have an impact.
My 2018 placement with WPREU went by in a flash and because of it, I am a far more well-rounded researcher with far more knowledge, skills and 'real-world' quantitative research experience. More than ever, I am certain that quantitative data analysis is the path I want to take for my future career - and the experience I have gained at WPREU will be invaluable to me in completing my MSc Data Science course to help me on my way to earning that career.
Why Numbers Matter
SMI's series on why it's important to get numbers right.
Explore the University
Discover what sets Sheffield apart at our next on-campus open day.