top of page

PhD Study Example: Preferred Virtual Assistant (VA) Gender in Self-Service 

Aim. Assess the impact of preferences for gender-matching with a VA on performance in a visual search task

My Role

Design and organise research

Work with designers

Liaise with Dundee Science Centre

Work with programmer

Conduct statistical analyses

Plan and conduct onsite interviews

Conduct qualitative analysis

Write up results

Report to NCR (funder)


Experimental design

Descriptive and inferential statistics

Qualitative analysis


Preferences for similarity-matching

Performance in visual search task

Problem Statement

My PhD applied psychological theory and research methods to human-computer interaction (HCI), examining the impact of a VA for use in self-service checkouts (SSCOs). An ongoing question throughout my studies was the impact of gender-matching on users, which is what I focus on here.



Visitors to Dundee Science Centre were invited to engage with a SSCO machine at a science exhibition ("The Fantasy and the Reality of Robots")

on which they could chose a VA from a selection of eight that differed by gender (male, female), realism (three-dimensional, cartoonized), and formality of dress (shirt, t-shirt). ​

After making a selection, participants engaged in a visual search task, which involved finding an item amongst a grid of 9 visually-similar items, twice when the VA looked in the direction of the grid, and twice when the VA looked straight ahead. ​

I conducted four-way log-linear analysis based on the study's hypotheses and found that, together, participant gender and age predicted gender-matching. Before this analysis, there seemed to be a lack of male preference for VA gender, even though male participants were slower to respond when a female VA was on screen.


Looking at participant age, it was actually the case that younger males tended to gender-match (until around 26-35 years old); this tendency steadily decreased with age. Meanwhile, regardless of age, females strongly preferred to gender-match with their VAs. This preference did not affect their reaction times in the visual search task.


These findings mimicked real life social interaction and relationship patterns based on sex and age, and suggested that the majority of users would prefer a female VA in a SSCO. 

3D Male VA


The findings contributed to a much larger investigation into VA gender and human-likeness, forming my PhD thesis and some publications.


This particular study, and versions of it, resulted in two publications: one in the International Society for Presence Research (2012), and one in Intelligent Virtual Agents (2013). 

My PhD research set me down the path of research into HCI, usable security, and, later, user experience, service design, and content writing.



There were so many! My PhD was one of the most prolonged stressful periods of my life. Looking back at the main challenges for this particular study, the most notable was an inability to control all variables. For example, I did not have time before the exhibition to measure perceptions of the VAs in a controlled environment. I mitigated this by spending some days at the exhibit to ask users their opinions on the VAs and why they chose them. I also could not control whether users participated more than once or truthfully reported their demographics. However, the large sample size limited the effect of noise in the data and the trends were very clear.

Analysing the data was not an easy task. I spent many weeks reading books on statistics and many more preparing the data, analysing it in SPSS, and justifying my choices, followed by a lengthy write-up period. The benefit of all this effort was that I found these choices easy to defend when it came to my Viva Voce in 2014. 

bottom of page