Along which identity lines does 21st-century Britain divide? Evidence from Big Brother

Abstract

This article measures discrimination in the reality TV show Big Brother, a high-stakes environment. Data on contestants’ nominations are taken from 35 series of the British version of the show, covering the years 2000–2016. Race and age discrimination are found, with contestants more likely to nominate those of a different race and those different in age from themselves. However, no discrimination is identified on the basis of gender, geographical region of origin, or level of education. Racial discrimination is driven by males, but females exhibit stronger age discrimination than males. Age discrimination is driven by the younger contestants discriminating against the older. Regional differences emerge, particularly between contestants from Greater London and those from the north of England; northerners have a stronger tendency to engage in racial and age discrimination, and to discriminate in favour of the opposite gender.

Keywords

Discrimination reality television social identity

Introduction

Because the impact of discrimination is considerable in any society, and the impact of extreme discrimination is catastrophic to the welfare of its victims, scholars from across the social sciences have long regarded discrimination as an interesting and important topic. However, not all discrimination is equal in strength. Discriminatory treatment occurs along multiple different identity lines, but the intensity of each of these types of discrimination varies across time and space. One country may commit genocide against racial minorities but enjoy a high level of gender equality, while another may deny women the vote but protect the elderly. A country where people of different races are forbidden from interacting may subsequently desegregate and liberalise its attitudes and behaviours. It is doubtful that findings on one form of discrimination can necessarily be generalised to other forms, or that findings on a particular form of discrimination within a particular population and era will hold true within other populations or eras. Therefore, it is important that researchers separately investigate different forms of discrimination, and that they do so with focus on specific populations and points in history.

This article focuses on discrimination in Britain, along multiple identity lines and within various sub-populations, in the 21st century. The country’s size and influential position in the world, as well as the historical context, make it a particularly compelling case to study. Prejudice and discrimination had existed overtly in the country in the 20th century, but by the arrival of the 21st century, Britain had taken on socially progressive governance, with a drive towards strong anti-discrimination legislation (Hepple et al., 2000). Certain forms of discrimination in Britain have become politically incorrect, or socially inappropriate (Barr et al., 2018). However, the backlash against such political correctness, exemplified in the rise of nationalism and the campaign to leave the European Union, suggests social identity and discrimination in Britain remain relevant, and make these important and timely matters for research.

In this study, discrimination is measured using data from a quasi-natural experiment: the British version of the reality TV show Big Brother, during the years 2000–2016. In Big Brother, a small group of contestants (housemates) live together in a house for a period of weeks or months with no contact with the outside world. Contestants are periodically evicted by a public vote, with those contestants facing the public vote determined by the prior nominations of each contestant. The last contestant to avoid eviction wins the show, and a large cash prize. Discrimination is measured through the nomination choices housemates make, by testing for the systematic tendency to nominate those with dissimilar identity characteristics.

Big Brother provides a useful environment in which to study discrimination. The mini-society contestants form in the house is – like in an experiment – closed from the external world, and alliances are allowed to develop naturally. However, the length and intensity of the time housemates are forced to spend together bring about an environment which would be difficult to create experimentally or observe naturally. The mundane environment – unlike in other reality TV shows or game shows, housemates do not need any special talents to progress through the show, but are required simply to keep living in a house – means that friendships are likely to form for similar reasons as they would in the outside world. At the same time, the stakes are high. The prize money for the show’s winner, along with the fame accruing to those surviving for longest, means that a housemate’s nominations – and any discrimination exhibited in them – have meaningful economic consequences in affecting nominees’ chances of experiencing these benefits. Discrimination in Big Brother can be measured cleanly, and the show’s longevity – with 35 series airing up to 2016 – allows the construction and analysis of a large data set.¹

I test for discrimination occurring on the basis of race, gender, age, geographical region of origin, and educational level. While some of these forms of discrimination have received a lot of quantitative analysis, others have received little (Rodgers, 2009). A vast body of empirical evidence documents the pervasiveness of racial discrimination across different countries and historical periods (Arrow, 1998; Zschirnt and Ruedin, 2016). Specifically in Britain, racial discrimination has been identified historically (e.g. Firth, 1981; Stewart, 1983) and more recently (Heath and Cheung, 2006; Wood et al., 2009).

As with race, there is substantial empirical evidence of gender discrimination across time and space, although also the suggestion that it has reduced in intensity over time (Jarrell and Stanley, 2004). Such discrimination has been measured within Britain (e.g. Wright and Ermisch, 1991), including in the 21st century (Joshi et al., 2007). The gender discrimination literature has tended to focus on the labour market, however; lab experiments, in contrast, have not often detected gender discrimination (Lane, 2016). It is therefore far from certain that one would expect to find gender discrimination on Big Brother.

Unlike race and gender, there has been relatively little quantitative research on age discrimination. The existing studies have, however, tended to show strong effects (e.g. Ahmed et al., 2012). This indeed has been the case specifically in 21st-century Britain, where a field experiment by Riach and Rich (2010) found levels of age discrimination in excess of those usually estimated for race.

Discrimination on the basis of geographical region has rarely been quantified. There is evidence of discrimination between residents of different neighbouring villages (e.g. Dugar and Shahriar, 2012), and between different nationalities (e.g. Carpenter and Cardenas, 2011); discrimination against rural-to-urban migrants in China has been documented (Kuang and Liu, 2012), while Heblich et al. (2015) found increased cooperation between Germans who speak the same regional accents. However, I am not aware of any research linking discrimination to geographical region within Britain. This is perhaps surprising; the importance of regional identity has been well-documented (Paasi, 2003) – particularly in Britain (Tomaney and Ward, 2000) – and regional parochialism can have negative consequences for a country’s cohesiveness.

There is also a lack of quantitative research into discrimination on the basis of educational level. Britain is a particularly interesting country in which to investigate this issue: George Orwell described England as ‘the most class-ridden country under the sun’ (Orwell, 1962: 10), but by the launch of Big Brother, the country’s prime minister, Tony Blair, had claimed that social class no longer existed.

Having measured discrimination along these various identity lines, this article furthermore investigates how the tendency to discriminate by race, gender and age differs according to a person’s demographic characteristics. Previous research supplies some evidence that, in general, males have a greater tendency for discriminatory behaviour than females (Balliet et al., 2014; Rand et al., 2009). This is backed up by an evolutionary ‘male warrior hypothesis’ and consistent evidence that males are more prone to prejudice against out-groups (McDonald et al., 2012). Evidence is lacking on age differences in discrimination, but older people have been found to be more prejudiced (e.g. Von Hippel et al., 2000). Specifically within Britain, strong age differences in support for anti-immigration political movements (Whitaker and Lynch, 2011) may lead one to expect racial discrimination to be stronger among older people.

Likewise, the purported concentration of a ‘liberal metropolitan elite’ – much discussed in the media following the 2016 EU referendum, in which there were strong regional and class-based differences in voting patterns – raises the possibility that the extent to which a person discriminates will depend both on which part of the country they are from and how highly educated they are, with particularly low discrimination expected among Londoners. However, in a field experiment, Riach and Rich (2010) found age discrimination to be stronger in London than the rest of England. Ultimately, little is known about demographic variation in the strength of discrimination in Britain, and these are therefore important empirical questions to address.

The main results of this study can be summarised as follows. Strong evidence of racial discrimination is found, with white contestants more likely to nominate non-white than white housemates. Strong evidence of age discrimination is also found: the likelihood of a housemate being nominated increases with their age difference from the nominator. However, I do not find evidence of discrimination against the opposite gender, nor of discrimination against those from different regional or educational backgrounds.

White males discriminate more strongly along racial lines than white females. However, females are more likely to discriminate by age. Regional differences also emerge, with the clearest differences between contestants from London and those from the north of England. Racial discrimination is stronger by (white) northerners than by (white) Londoners, and age discrimination is stronger by northerners than by Londoners, but Londoners are more likely to discriminate against the opposite gender, with a weakly significant tendency among northerners to nominate their own gender. The strongest effect of age difference appears on the nomination choices of younger contestants, indicating that the age discrimination is driven primarily by negative treatment towards elder housemates by the younger. No type of discrimination significantly differs between more and less highly educated individuals.

As well as contributing to the literature on discrimination in general, and in 21st-century Britain specifically, this article also adds to the growing literature analysing discrimination by contestants on game shows and reality TV. Studies on game shows have found contestants discriminating on the basis of race (Anwar, 2012), gender (Atanasov and Dana, 2013), age (List, 2006) and beauty (Belot et al., 2012; Hamermesh, 2012), although others have found little evidence of discrimination (Antonovics et al., 2005; Levitt, 2004). In analyses of the reality TV show Come Dine With Me, Ahmed (2013) and Schüller et al. (2014) identified contestant behaviour consistent with ethnic discrimination. The two most similar papers to this study both measure discrimination on Survivor, another reality show whose format involves contestants voting to eliminate one another; Dilks et al. (2010) and Wall (2011) both find discrimination against women on the show, while the former also finds discrimination against minorities and older contestants.² The setting explored in this study differs from Survivor both in the population from which contestants are drawn (Britain rather than the United States) and, to some extent, in the nature of the contest.³

Background on Big Brother

Big Brother is a reality TV show with a game show–like format. While many different versions have been produced internationally, this article concentrates solely on the British show, which has run since 2000. Although the format is subject to occasional twists (see the ‘Data and empirical strategy’ section), in general, it proceeds as follows. In each series of the show, contestants live together in a house for a period of weeks or months, filmed constantly, unable to leave and with no contact with the outside world. At regular intervals – usually once a week – each housemate is required to nominate a specified number of others (normally two) to face eviction. The housemates who receive the most nominations (generally those finishing equal-second or higher in the nominations rankings, although this threshold is sometimes lowered) subsequently face a public vote. Of these, the person – or, in occasional twists, more than one person – receiving the lowest public support to remain in the house is evicted. Usually, nominations are made in secret and housemates are forbidden from discussing among themselves who they have nominated or plan to nominate; although all housemates are told in advance which contestants will face the public vote for a given eviction, they are not informed how many nominations each received or who nominated whom. These secrecy rules have, however, sometimes been suspended. In the final round of the series, no nominations take place and all housemates face the public vote, with the one receiving the most votes deemed the winner and scooping a large cash prize, typically around £100,000.

In addition to the 17 regular series of Big Brother that aired up to the end of the 2016, there were also 18 series with celebrity contestants. The basic format remains unchanged between the two versions, although the celebrity series are shorter, with more frequent nominations and evictions. Celebrities receive participation fees, and the winner’s cash prize is donated to a charity of their choice. The environment remains one of high stakes, however; surviving in the show for longer, and ultimately winning, boosts a celebrity’s fame and image, which is likely to lead to substantial future financial gain. This study uses data from both the regular and celebrity series.

There are, therefore, material consequences to any discrimination in contestants’ nominations in both the regular and celebrity versions of the show. Nominating an individual makes them more likely to face the next public vote, thereby reducing their chance of winning the cash prize (in the regular show) and also restricting their likelihood of having a long run in the house and enjoying the associated benefits of fame and, potentially, enhanced self-esteem (in both versions of the show). If particular groups encounter disproportionately high nomination rates, this places them at a material disadvantage.

The main concerns about using game shows and reality TV for behavioural research relate to the representativeness of contestants and, therefore, the external generalisability of findings. Table 1 addresses contestants’ demographic representativeness. The nomination observations within this article’s data set are broken down according to the demographics of the contestants making them (presenting proportions of observations, rather than of contestants, accounts for the fact that types of contestants who survive in the show for more rounds are more strongly represented in the data⁴). Where available, data from the 2011 UK census (UK Census, 2011) are also presented for these demographic variables. Like many reality programmes, Big Brother disproportionately features young adults, although in the celebrity series, this is barely the case. In other respects, however, contestants are fairly demographically representative of Britain; males contribute just over half the nomination choices in the data set, and whites contribute around four-fifths. In the regular series, contestants represent diverse geographical backgrounds that are reasonably nationally representative, although Londoners are more strongly represented and other southerners less strongly represented than in the wider population (this is not coded for the celebrity series).⁵

Table 1.

Breakdown of nomination choices in data set by demographics of nominators.

Variables	Regular series			Celebrity series			Britain (2011 census)
Variables	Mean (SD)	Minimum	Maximum	Mean (SD)	Minimum	Maximum	Mean (SD)	Minimum	Maximum
Male	0.543	0	1	0.505	0	1	0.491	0	1
White	0.769	0	1	0.836	0	1	0.872	0	1
Age	26.4 (6.93)	18	59	38.4 (13.2)	18	82	39.5 (N/A)	0	90+
From Greater London	0.250	0	1				0.129	0	1
From non-London South	0.196	0	1				0.313	0	1
From Midlands	0.118	0	1				0.160	0	1
From North	0.237	0	1				0.236	0	1
From outside England	0.196	0	1				0.161	0	1
Higher education	0.386	0	1				N/A

Note: See ‘Independent variables’ section for definitions of the variables. Regional and class variables are not coded for celebrity series.

Although the above suggests that contestants are fairly representative on certain observable characteristics, unobservables may be a different matter. Indeed, certain personality types are very likely to be overrepresented on Big Brother. Most obviously, those prepared to place themselves in such an environment will tend to be disproportionately extroverted, exhibitionistic and in search of fame. Although little research has been conducted on the relationship between discrimination and personality type, what evidence does exist would not support a clear argument that Big Brother contestants should be more or less discriminatory than the wider population on the basis of the aforementioned traits: Corr et al. (2015) found no consistent association between extroversion and discrimination; Ekehammar and Akrami (2003) also found extroversion to be uncorrelated with prejudice. Big Brother contestants are also likely to be disproportionately competitive; it is theoretically possible, though empirically uncertain, that more competitive personalities may be associated with inflated tendencies to discriminate (see, for example, McDonald et al., 2012).

Conceivably, the producers’ decisions over who to accept on the show could skew the levels of particular types of discrimination. For example, they could handpick racially or sexually prejudiced applicants, or applicants representing negative stereotypes associated with particular identity groups, if such individuals were thought to be a good source of entertainment. However, as I will show in the subsection ‘Does behaviour differ between regular and celebrity series?’, similarities between the behaviour of regular and celebrity series contestants cast doubt over such an interpretation.

Data and empirical strategy

Nomination data were taken from Wikia (2017), an entertainment fandom website, with the exception of the tenth celebrity series, for which it was missing. For this series, the data are taken from Wikipedia (2017). As a precautionary measure resulting from using these potentially unreliable sources, I cross-checked 200 nominations from 10 series by watching online footage of them.⁶ On no occasion did I find inconsistencies with my data set, which suggests it is of a reliable nature.

The data set comprises information on 1919 sets of nomination choices by 461 contestants from across the 35 series. Within each set of nomination choices, there is one observation for each housemate who the nominator could potentially nominate. The dependent variable, Nominated, is binary: it takes the value of 1 if the housemate is nominated by the given nominator, and 0 otherwise. In total, there are 17,449 observations in the data set.

In Big Brother, twists are occasionally introduced which temporarily alter the format of nominations and evictions as described in the ‘Background on Big Brother’ section. Certain housemates are on occasion prohibited from casting nominations, or from being nominated – for example, it is occasionally decreed that a housemate will automatically face the public vote and therefore cannot be nominated. The data set takes this into account; within each set of nominations, an observation is only recorded for each housemate who is eligible to be selected by the nominator in question.

I exclude data on some nomination choices where a format twist alters either the consequences of a nomination (at least as understood by the nominator) or the nature of the decision-making agent choosing it in such a way that the setting is not meaningfully comparable to that from which the rest of the data is drawn. I exclude the following:

Cases where nominations are made by a friend or relative on behalf of a housemate.

Cases where nominations are made by a temporary houseguest, who is not eligible to win the show, or by a housemate who has already been evicted and therefore cannot win.

Cases where nominations are made by the group decisions of two or more housemates (the exception to this is where a pair of twins competes as a single contestant, in which case they are treated as one housemate in the data).

Cases where housemates are required to make ‘nominations’, but the nomination relates to something other than eviction, and housemates are aware that this is the case (for instance, in the first week of the fifth regular series housemates nominated others to lose access to their belongings).

One case where one housemate had been forewarned about a twist which would lead to the contestant who received the most nominations in that round of voting being the only housemate exempt from facing the public vote. In this case, nomination decisions by and relating to this particular housemate are excluded.

Cases where nominations are made unwittingly (for instance, if Big Brother decrees that a housemate will be considered to have nominated the first person they touch on a particular day, but does not inform the housemate of this).

Cases where a format twist takes place and the data source does not provide sufficient information on the twist as to make clear which housemates were eligible to be nominated by which others.

Still included in the data, however, are the following:

Cases where a twist will lead to a set of nominations having non-standard consequences, but housemates are not aware of this at the time of making their decisions (for instance, twists have been introduced where the nominated housemates will face a public vote over whom to move to a secret area within the house for a designated period of time, rather than over whom to evict).

Cases where housemates nominate which others they want to stay in the house, as opposed to which they want to face eviction. In this case, Nominated takes the value of 1 for those not nominated to stay, and 0 for those nominated to stay.

Cases where a nomination has a greater amount of power than usual. This includes cases where housemates vote on whom to directly evict from the house, skipping the public vote stage. It also includes cases where one housemate is given the power to select another who will automatically face the public vote, regardless of how many other nominations they receive, or even who will automatically face the public vote in every subsequent week until they are evicted.

Cases where a nomination must be administered in a non-standard way (e.g. by the nominator touching the nominee, rather than announcing their name), and nominators are made aware of this.

Each housemate’s decision over whom to nominate on a given occasion takes the form of a discrete choice problem. Each housemate must select between alternatives (i.e. potential nominees) where their utility from selecting a particular alternative depends upon its attributes (i.e. the potential nominee’s personal characteristics) and these attributes vary between alternatives. As such, conditional logit regressions represent the appropriate modelling technique. The conditional logit estimates the effects that the attributes of an alternative have on the probability of this alternative being selected from among a given set. In the present case, it estimates the effects that a potential nominee’s personal characteristics have on the probability of them being nominated from the set of potential nominees available in a given nomination decision. The results of the conditional logit are not biased by differences between rounds in the number of individuals a housemate must nominate, nor in the number of potential nominees, nor in the proportion of a particular type (e.g. race, gender) of potential nominees, as these variables are held constant across all observations within a particular set of nomination choices. Standard errors in the models are corrected for heteroscedasticity, with each series treated as providing a cluster of observations.

Independent variables

Data are available from Wikia on housemates’ age, home place and occupation (for the tenth celebrity series, data on age are missing from Wikia, so it is taken from Wikipedia instead). This information is submitted by the contestants themselves and broadcast when they enter the house. I used photos on Wikia to code housemates’ gender and race. The models I estimate include the following independent variables.

Different gender

A dummy variable equal to 1 if the potential nominee shares the same gender as the nominator, and equal to 0 otherwise.⁷

Age difference

The modulus of the distance in years between the potential nominee’s age and the nominator’s age.

Non-white

A dummy variable taking the value of 1 if a potential nominee is non-white and 0 if they are white. It is generally straightforward to observe which housemates are white, although, to minimise any possible biases in such identifications, I coded this variable before collecting the nominations data in order to be able to make borderline decisions blind to the effects they would have on my results. I did not seek to code the race of non-white housemates. The potential nominee is non-white in only 21.4% of cases, and non-white contestants represent a wide variety of different racial groups, with many being mixed race, which would often make coding this variable rather tricky. When analysing the effect of race difference, I will therefore drop nominations made by non-white housemates. On the reduced sample, the non-white variable will then represent race difference between nominator and potential nominee.

Different educational level

I split the sample into high and low educational levels, according to whether or not housemates’ self-recorded occupations usually require university education. Current university students are placed in the high educational level. I code this variable only for contestants in the regular series, as the nature of many celebrities’ self-reported occupations (usually in the entertainment industry) makes it unlikely that this variable could be coded for them with any reliability. Different educational level is a dummy variable equal to 1 if the potential nominee’s educational level is different from that of the nominator, and equal to 0 if it is the same.

Different region

I group housemates into geographic regions, according to their self-identified home place. Housemates are categorised as being from the north of England; the midlands; Greater London; the rest of the south of England, excluding Greater London; or from outside England. The English regions are based on definitions in the 2011 UK census: the north of England contains the areas defined as North East, North West, and Yorkshire and the Humber; the midlands contain the areas defined as East Midlands and West Midlands; the south, excluding London, contains the areas defined as the South East, South West and East. Different region takes the value of 1 if the potential nominee is categorised as being from a different region from the nominator, and 0 if they are categorised as being from the same region. When analysing this variable, I will drop nominations made by housemates from outside England, as they do not represent a coherent regional in-group.⁸ I do not code this variable for celebrities as it is unlikely to reflect a strong sense of identity. For instance, a high proportion of the celebrities have listed as their home place somewhere within Greater London or Los Angeles, where the British and American entertainment industries are concentrated, but it is likely many of these people originate from elsewhere.

Results

General patterns of discrimination

Table 2 presents the results of conditional logit regressions run on the dependent variable Nominated. Odds ratios on the independent variables indicate the effects these variables have on the odds of an individual being nominated from within a given set of available nominees (where the odds represent the ratio of the probability of being nominated to the probability of not being nominated). An odds ratio above 1 means a variable increases the odds (and therefore the likelihood) of one being nominated, while an odds ratio below 1 means it reduces it. Model (1) is run on the entire data set. An odds ratio of 1.086 on different gender implies that, across all series, potential nominees whose gender differs from that of the contestant nominating have higher odds of being nominated than those whose gender is the same by a factor of 1.086. However, this variable does not reach conventional levels of statistical significance, so it cannot be claimed that any gender discrimination is evident in the data.

Table 2.

Conditional logit regressions on the odds of being nominated.

Dependent variable: nominated
	(1)	(2)	(3)	(4)
	Whole sample	White contestants	Regular series	Regular series – English
Different gender	1.086 (0.0567)	1.066 (0.0564)	1.036 (0.0685)	1.005 (0.0584)
Age difference	1.023*** (0.00469)	1.022*** (0.00521)	1.026*** (0.00662)	1.023*** (0.00565)
Non-white	1.271*** (0.104)	1.313*** (0.115)	1.351*** (0.121)	1.380*** (0.114)
Different education level			1.136 (0.103)	1.159 (0.106)
Different region				1.073 (0.105)
Pseudo R²	0.0086	0.0088	0.0088	0.0090
Observations	17,449	13,768	12,135	9751

Note: Table 2 presents conditional logit models, with odds ratios presented. Note that an odds ratio of more than 1 indicates a positive effect, while an odds ratio of less than 1 indicates a negative effect. The dependent variable indicates whether or not an individual is nominated by a given contestant in a given round. Model (1) includes the nomination choices of all contestants; Model (2) includes only the nomination choices of white contestants; Model (3) includes only the nomination choices of regular series contestants; Model (4) includes only the nomination choices of regular series contestants whose home region is in England. Standard errors, in parentheses, are corrected for heteroscedasticity, clustering by series.

p < 0.1; **p < 0.05; ***p < 0.01.

In contrast, age difference is significant at the 1% level, strongly suggesting the likelihood of an individual being nominated is greater the larger their distance in age from the contestant nominating. An odds ratio of 1.023 implies that a one-year increase in age difference is associated with a 2.3% increase in the odds of being nominated. Assuming linearity, the model implies, for instance, that a contestant with 10-year difference in age from the person nominating has 1.255 times higher odds of being nominated, and a contestant with 20-year difference in age has 1.576 times higher odds of being nominated, than another contestant who is the same age as the person nominating.

Model (1) also estimates non-white to be significant at the 1% level, demonstrating that being non-white increases a housemate’s likelihood of being nominated. While this represents discrimination in general, the specific effect of race difference on nomination choices is estimated by Model (2), which excludes all data on nominations by non-white contestants and thus estimates how much more likely white housemates are to nominate non-whites than whites. The model finds race to be a highly significant explanatory variable; a non-white person has 1.313 times higher odds than a white person of being nominated by a white contestant.⁹

The effects of differences in educational level and home region on nomination choices are estimated using data only from the regular Big Brother series. Model (3) estimates a positive but insignificant odds ratio on different education level, indicating that nominators are not significantly more likely to nominate contestants from a different educational class than those from their own. Model (4) investigates the effects of regional backgrounds. It excludes data from the nominations of non-English housemates, whose backgrounds are disparate and not coded as belonging to specific regions.¹⁰ The variable different region is positive but insignificant, suggesting that sharing a housemate’s regional background does not make an individual significantly less likely to be nominated by them. There is therefore no significant evidence of regional or educational discrimination. This may in part be due to testing them using smaller sample sizes, but note that the odds ratios on different education level and different region are much closer to 1 than those on non-white.

Analysis of racial discrimination within different white sub-groups

I now consider how levels of racial discrimination differ between different white sub-groups. I re-run the conditional logit regressions, with different gender, age difference and non-white as independent variables, on specific subsets of the data representing the nomination choices of particular demographic groups. I separately consider male and female nominators; nominators aged 18–25, 26–40 and 41+; nominators from London, the midlands, the non-London south, and the north; and nominators coming from high and low educational levels. In all cases, the sample excludes non-white nominators; for the regional and education-based groups, it also excludes nominators from the celebrity series. Table 3 reports the odds ratio estimated on the non-white variable by the regression focusing on the behaviour of each sub-group. The significance of differences between the odds ratios estimated for different sub-groups is calculated by running pooled models with interaction terms; where differences are found to be significant, this is indicated by connecting lines at the right hand side of the table. The full output of all regressions is presented in Online Appendix C.

Table 3.

Racial discrimination in nominations by sub-groups of white contestants.

Sub-group	Odds ratio on non-white
Gender
Male
Female
Age range
18–25
26–40
41+
Region
London
Midlands
South (non-London)
North
Educational level
High educational level
Low educational level

Note: Table 3 presents the effect of being non-white on the odds of being nominated by contestants belonging to various white sub-groups, as estimated by conditional logit regressions including data only from the choices of members of the applicable groups. The models run on regional and social class sub-groups exclude data from celebrity series. Note that an odds ratio of more than 1 indicates a positive effect, while an odds ratio of less than 1 indicates a negative effect. Significance of differences between the odds ratios of different models are estimated by running pooled models with interaction terms. Standard errors, in parentheses, are corrected for heteroscedasticity, clustering by series.

p < 0.1; **p < 0.05; ***p < 0.01.

Table 3 shows that the racial discrimination exhibited by white males is strongly significant, while for white females it is not significant. This difference between genders is found to be significant at the 5% level. Stronger racial discrimination is found by white housemates aged 18–25 than those aged 26–40 or 41+ (with the discrimination exhibited by the latter group not significant), but none of the differences in behaviour between any of these groups are estimated as significant. Regarding region, white housemates from London and the midlands display relatively weak and insignificant racial discrimination (though the insignificance may in part be due to sample size); the odds ratios are higher and significant for housemates from the non-London south and the north. Racial discrimination by white northerners is stronger than that of white Londoners, significant at the 5% level. Regarding educational level, whites of both high and low educational levels discriminate significantly against non-whites, and the difference between the two groups is not significant.

Analysis of gender and age discrimination within different sub-groups

I replicate the approach of the previous subsection to investigate how levels of gender and age discrimination differ between sub-groups. The same regressions are run, with different gender, age difference and non-white as independent variables, but now the nomination choices of non-white housemates are also included in the data. Table 4 reports the odds ratios estimated on different gender (on the left hand side) and age difference (on the right hand side) in each regression. As above, pooled models with interaction terms are run to investigate whether the odds ratios estimated for separate sub-groups are significantly different, with connecting lines to the right of the odds ratios indicating where this is the case. The full output of all regressions is again presented in Online Appendix C.

Table 4.

Gender and age discrimination in nominations by sub-groups of contestants.

Sub-group	Odds ratio on different gender	Odds ratio on age difference
Gender
Male
Female
Age range
18–25
26–40
41+
Region
London
Midlands
South (non-London)
North
Educational level
High educational level
Low educational level

Note: Table 4 presents the effect of gender and age difference on the odds of being nominated by contestants belonging to various sub-groups, as estimated by conditional logit regressions including data only from the choices of members of the applicable groups. The models run on regional and social class sub-groups exclude data from celebrity series. Note that an odds ratio of more than 1 indicates a positive effect, while an odds ratio of less than 1 indicates a negative effect. Significance of differences between the odds ratios of different models are estimated by running pooled models with interaction terms. Standard errors, in parentheses, are corrected for heteroscedasticity, clustering by series.

p < 0.1; **p < 0.05; ***p < 0.01.

The left hand side of Table 4 shows there are few significant differences between different sub-groups in terms of gender discrimination, and most sub-groups are not found to significantly favour either their own or the opposite gender. However, strong regional differences do emerge. Northerners are found to exhibit opposite-gender favouritism – they are more likely to nominate housemates of their own gender than of the other – and, in this respect, their behaviour differs from that of Londoners and other southerners, both at the 1% level.

While strongly significant discrimination on the basis of age difference is found among both males and females, it is found to be significantly stronger among females. Important differences between age groups also emerge. While 18- to 25-year-olds and 26- to 40-year-olds significantly discriminate on the basis of age difference, those aged 41 and above do not; in this respect, the behaviour of the 41+ group is found to significantly differ from that of the two other age groups. This suggests the age-based discrimination found across the whole sample is driven primarily by the behaviour of the younger housemates towards the older.¹¹ Significant discrimination on the basis of age difference is found among housemates from the north and midlands, but not among those from London or the rest of the south: the difference in behaviour between northerners and Londoners is significant at the 5% level. Housemates of high and low educational levels both significantly discriminate on the basis of age difference, and do not significantly differ from each other in this behaviour.

Does behaviour differ between regular and celebrity series?

In the ‘Background on Big Brother’ section, I discuss the issue of how generalisable behaviour in game shows is to the external world. Here, I offer some evidence in favour of the generalisability of my results, through investigating whether discriminatory behaviour differs between the regular and celebrity versions of Big Brother. If levels of discrimination on the show were the product of the housemate selection process, one would expect differences to appear between the versions, as each operates very different procedures for choosing its contestants. For every regular series, thousands of members of the public apply and the show’s producers are able to select which small number are successful. For the celebrity series, in contrast, the producers invite and offer to pay famous people to participate, and these individuals decide whether or not to accept the invitations; judging by the very obscure nature of many of the contestants’ celebrity, it is reasonable to assume that many rejections from more famous invitees occur. Therefore, while it may be possible for the producers to bias the selection process of the regular series to lead to particular levels of discrimination (for instance, by handpicking racially or sexually prejudiced applicants, or applicants representing negative stereotypes associated with particular identity groups, in order to create entertaining television), it is unlikely that the same would occur in the celebrity series.

I estimate whether discriminatory behaviour differs between the regular and celebrity series by re-running the conditional logit regression on the whole sample, and including interaction terms. These results are presented in Table 5. The variables non-white, different gender and age difference, respectively, represent the levels of discrimination nominators display against housemates who are non-white, of the opposite gender, or of a different age to them, in the regular series. The interaction terms between celebrity and these variables indicate how much stronger or weaker such levels of discrimination are in the celebrity series. For instance, the odds ratio of 1.163 on Celebrity × Different Gender indicates the disadvantage of being of a different gender to the nominator relative to being of the same gender, in terms of the odds of being nominated, is 1.163 times stronger in the celebrity series than in the regular series. None of the interaction terms have odds ratios which reach conventional levels of significance. A test of joint significance on the three interaction terms also returns a null result (p = 0.219), confirming that the hypothesis of discriminatory behaviour being indistinguishable between celebrity and regular series housemates cannot be rejected.

Table 5.

Conditional logit regression on odds of being nominated.

Dependent variable: nominated
	Whole sample
Non-white	1.339*** (0.115)
Different gender	1.036 (0.0671)
Age difference	1.026*** (0.00653)
Celebrity × Non-White	0.810 (0.181)
Celebrity × Different Gender	1.163 (0.113)
Celebrity × Age Difference	0.996 (0.00911)
Pseudo R²	0.0093
Observations	17,449

Note: Table 5 presents a conditional logit model, with odds ratios presented. Note that an odds ratio of more than 1 indicates a positive effect, while an odds ratio of less than 1 indicates a negative effect. The dependent variable indicates whether or not an individual is nominated by a given contestant in a given round. Standard errors, in parentheses, are corrected for heteroscedasticity, clustering by series.

p < 0.1; **p < 0.05; ***p < 0.01.

A further robustness check is presented in Online Appendix B, where a similar approach is used to test whether housemates’ nomination behaviour significantly differs between the years 2000–2010 and 2011–2016. Big Brother moved from Channel 4 to Channel 5 in 2011, and certain aspects of the show changed noticeably (for instance, the daily highlights show introduced a more over-the-top narrative style). It is therefore plausible that there were also changes in producers’ preferences over contestant types. However, Model (3) in Online Table B1 shows the levels of race, gender and age discrimination do not significantly differ between the two periods. A joint significance test on the three interactions terms returns a null result (p = 0.263).

Conclusion

This article has measured discrimination in the nomination choices of contestants on the British version of Big Brother. Discrimination is found on the basis of race and age, but not gender, region of origin or education level.¹² A 10-year age difference between two housemates results in a similar level of discrimination as does the two belonging to different races. Given the high stakes, the economic consequences of such discrimination are important; ethnic minorities and old people are nominated more often and are therefore ceteris paribus likely to be eliminated earlier, reducing their expected material returns from participating on the show.

These results mirror previous research showing strong economic discrimination occurring on the basis of race (e.g. Zschirnt and Ruedin, 2016) and age (e.g. Riach and Rich, 2010), and relatively weak gender discrimination occurring in non-labour-market contexts (Lane, 2016). Focusing specifically on Britain, the results suggest that after decades of non-white immigration, race still remains a defining identity characteristic in the country. However, individuals do not appear to have a clear tendency to favour those from the same geographical region or educational background, suggesting that regional and class identities are now relatively weak in Britain. This may be a reflection of levels of geographical and social mobility in the country being higher today than at times during the past.

Various demographic differences emerge from the data. Males are more prone to racial discrimination than females, which supports the ‘male warrior hypothesis’ that men are more hostile than women to outsiders (McDonald et al., 2012). Conversely, however, females have a greater tendency to discriminate on the basis of age difference. The regional differences which emerge are striking at times; Londoners and northerners significantly differ in each of the three respects that they are compared. Londoners do not significantly engage in either racial or age discrimination, whereas northerners strongly engage in both. This, arguably, chimes with the ‘liberal metropolitan elite’ concept. Meanwhile, northerners have a significantly greater tendency than Londoners for favouritism towards the opposite gender.

However, no differences emerge between the tendencies of more or less highly educated individuals. Findings on age are also not necessarily what one would expect. Where it might be expected that older contestants would display more racial discrimination, it is in fact measured to be stronger among the youngest, although the difference is not significant. Younger housemates are also significantly more likely to discriminate on the basis of age difference than older ones are. It is worth noting that older housemates are relatively rare, so the discrimination they face from the younger may be in part related to their minority status in the house.

This article is not able to confirm the underlying motivations beneath the various forms of discrimination analysed herein. One possibility is that discrimination in Big Brother is largely the consequence of homophily, the tendency for people to associate with and form positive relationships with those who share similar characteristics (see, for example, McPherson et al., 2001). This might be the case because in Big Brother – unlike other reality TV shows like Survivor – contestants’ well-being during their time in the house does not depend on the practical skills of others. In such a mundane setting, they may be likely to nominate primarily along friendship lines. If so, the results would suggest the presence of strong homophily on the basis of race and age, but not on the basis of the other identity types.¹³ It would mean that the economic consequences of the racial and age discrimination among Big Brother housemates (exclusion from the material benefits of prize money and fame) are the result of exclusion of racial minorities and older people from the friendship groups of the white, young majority of contestants. These are not dissimilar from the economic consequences of discrimination resulting from homophily in the outside world: if, for instance, nepotism towards friends is made in hiring decisions, the result can be the loss of earnings for out-groups excluded from homophilous friendship networks.

Somewhat relatedly, it should be noted that romantic relationships (actual or hoped for) may motivate discrimination along some identity lines. In particular, romantic interest may partly explain why we do not see a particular tendency to nominate those of the opposite gender. A lack of romantic interest may also partly explain the tendency for the younger housemates to nominate the older.

However, beyond friendships and romantic interests, other strategic criteria for nominations cannot be excluded. Moreover, whether or not we consider the discrimination in Big Brother to be driven by homophily, we are not able to discern which of the two traditional categories of economic discrimination – taste-based or statistical discrimination – to label it under. For instance, the white housemates may have developed friendships with other white housemates either because they had an intrinsic preference for doing so, or because they saw it as somehow advantageous towards their probability of being victorious.

Supplemental Material

big_brother_nominations_combined – Supplemental material for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother

Supplemental material, big_brother_nominations_combined for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother by Tom Lane in Rationality and Society

Supplemental Material

Big_Brother_nominations_combined_data_key – Supplemental material for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother

Supplemental material, Big_Brother_nominations_combined_data_key for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother by Tom Lane in Rationality and Society

Supplemental Material

Lane2020_Rationality_and_Society_do_file – Supplemental material for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother

Supplemental material, Lane2020_Rationality_and_Society_do_file for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother by Tom Lane in Rationality and Society

Supplemental Material

Online_Appendix_and_tables – Supplemental material for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother

Supplemental material, Online_Appendix_and_tables for Along which identity lines does 21st-century Britain divide? Evidence from Big Brother by Tom Lane in Rationality and Society

Footnotes

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

ORCID iD

Tom Lane

Supplementary material

Supplemental material for this article is available online.

Notes

References

Ahmed

(2013) Co-ethnic preferences in a cooking game: a study based on come dine with me in Sweden. Ethnic and Racial Studies 36: 2220–2236.

Ahmed

Andersson

Hammarstedt

(2012) Does age matter for employability? A field experiment on ageism in the Swedish labour market. Applied Economics Letters 19: 403–406.

Antonovics

Arcidiacono

Walsh

(2005) Games and discrimination lessons from the weakest link. Journal of Human Resources 40: 918–947.

Anwar

(2012) Testing for discrimination: evidence from the game show Street Smarts. Journal of Economic Behavior & Organization 81: 268–285.

Arrow

(1998) What has economics to say about racial discrimination? The Journal of Economic Perspectives 12: 91–100.

Atanasov

Dana

(2013) The price is sexist: taste based discrimination by contestants on the price is right. Available at: https://pdfs.semanticscholar.org/f343/34534babde9693261fc69f65f1ba4d8d7d03.pdf

Balliet

de Dreu

CKW

(2014) In-group favoritism in cooperation: a meta-analysis. Psychological Bulletin 140: 1556–1581.

Baltussen

van den Assem

van Dolder

(2016) Risky choice in the limelight. Review of Economics and Statistics 98: 318–332.

Barr

Lane

Nosenzo

(2018) On the social inappropriateness of discrimination. Journal of Public Economics 164: 153–164.

10.

Beetsma

Schotman

(2001) Measuring risk attitudes in a natural experiment: data from the television game show Lingo. The Economic Journal 111: 821–848.

11.

Belot

Bhaskar

van de Ven

(2010) Promises and cooperation: evidence from a TV game show. Journal of Economic Behavior & Organization 73: 396–405.

12.

Belot

Bhaskar

Van De Ven

(2012) Beauty and the sources of discrimination. Journal of Human Resources 47: 851–872.

13.

Blavatskyy

Pogrebna

(2008) Risk aversion when gains are likely and unlikely: evidence from a natural experiment with large stakes. Theory and Decision 64: 395–420.

14.

Carpenter

Cardenas

(2011) An intercultural examination of cooperation in the commons. Journal of Conflict Resolution 55: 632–651.

15.

Corr

Heap

SPH

Seger

, et al. (2015) An experiment on individual ‘parochial altruism’ revealing no connection between individual ‘altruism’ and individual ‘parochialism’. Frontiers in Psychology 6: 1261.

16.

Deck

Lee

Reyes

(2008) Risk attitudes in large stake gambles: evidence from a game show. Applied Economics 40: 41–52.

17.

Dilks

Thye

Taylor

(2010) Socializing economic theories of discrimination: lessons from Survivor. Social Science Research 39: 1164–1180.

18.

Dugar

Shahriar

(2012) Group identity and the moral hazard problem: experimental evidence. Journal of Economics & Management Strategy 21: 1061–1081.

19.

Ekehammar

Akrami

(2003) The relation between personality and prejudice: a variable-and a person-centred approach. European Journal of Personality 17: 449–464.

20.

Firth

(1981) Racial discrimination in the British labor market. Industrial & Labor Relations Review 34: 265–272.

21.

Haan

Los

Riyanto

(2011) Signaling strength? An analysis of decision making in the weakest link. Theory and Decision 71: 519–537.

22.

Hamermesh

(2012) Tall or taller, pretty or prettier: is discrimination absolute or relative? IZA Journal of Labor Economics 1: 2.

23.

Heath

Cheung

(2006) Ethnic penalties in the labour market: employers and discrimination. Available at: http://orca.cf.ac.uk/56299/

24.

Heblich

Lameli

Riener

(2015) The effect of perceived regional accents on individual economic behavior: a lab experiment on linguistic performance, cognitive ratings and economic decisions. PLoS ONE 10: e0113475.

25.

Hepple

Coussey

Choudhury

(2000) Equality: A New Framework: Report of the Independent Review of the Enforcement of UK Anti-Discrimination Legislation. Oxford: Hart Publishing.

26.

Hogarth

Karelaia

Trujillo

(2012) When should I quit? Gender differences in exiting competitions. Journal of Economic Behavior & Organization 83: 136–150.

27.

Jarrell

Stanley

(2004) Declining bias and gender wage discrimination? A meta-regression analysis. Journal of Human Resources 39: 828–838.

28.

Joshi

Makepeace

Dolton

(2007) More or less unequal? Evidence on the pay of men and women from the British birth cohort studies. Gender, Work & Organization 14: 37–55.

29.

Keldenich

Klemm

(2014) Double or nothing?! Small groups making decisions under risk in ‘Quiz Taxi’. Theory and Decision 77: 243–274.

30.

Kuang

Liu

(2012) Discrimination against rural-to-urban migrants: the role of the Hukou system in China. PLoS ONE 7: e46932.

31.

Lane

(2016) Discrimination in the laboratory: a meta-analysis of economics experiments. European Economic Review 90: 375–402.

32.

Lane

(2019) Get her off my screen: taste-based discrimination in a high-stakes popularity contest. Oxford Economic Papers 71(3): 548–563.

33.

Lee

(2009) American idol: evidence on same-race preferences. The BE Journal of Economic Analysis and Policy 9: 1–21.

34.

Levitt

(2004) Testing theories of discrimination: evidence from weakest link. The Journal of Law and Economics 47: 431–452.

35.

Lindquist

Säve-Söderbergh

(2011) ‘Girls will be girls’, especially among boys: risk-taking in the ‘daily double’ on jeopardy. Economics Letters 112: 158–160.

36.

List

(2006) Friend or foe? A natural experiment of the prisoner’s dilemma. The Review of Economics and Statistics 88: 463–471.

37.

McDonald

Navarrete

Van Vugt

(2012) Evolution and the psychology of intergroup conflict: the male warrior hypothesis. Philosophical Transactions: Biological Sciences 367: 670–679.

38.

McPherson

Smith-Lovin

Cook

(2001) Birds of a feather: homophily in social networks. Annual Review of Sociology 27: 415–444.

39.

Orwell

(1962) Socialism and the English Genius. London: Secker & Warburg.

40.

Paasi

(2003) Region and place: regional identity in question. Progress in Human Geography 27: 475–485.

41.

Pogrebna

(2008) Naïve advice when half a million is at stake. Economics Letters 98: 148–154.

42.

Pogrebna

Blavatskyy

(2009) Coordination, focal points and voting in strategic situations: a natural experiment. Public Choice 140: 125–143.

43.

Post

Van den Assem

Baltussen

, et al. (2008) Deal or no deal? Decision making under risk in a large-payoff game show. The American Economic Review 98: 38–71.

44.

Rand

Pfeiffer

Dreber

, et al. (2009) Dynamic remodeling of in-group bias during the 2008 presidential election. Proceedings of the National Academy of Sciences of the United States of America 106: 6187–6191.

45.

Riach

Rich

(2010) An experimental investigation of age discrimination in the English labor market. Annals of Economics and Statistics/Annales d’économie et de Statistique: 169–185.

46.

Rodgers

(2009) Handbook on the Economics of Discrimination. Cheltenham: Edward Elgar Publishing.

47.

Schüller

Tauchmann

Upmann

, et al. (2014) Pro-social behavior in the TV show ‘come dine with me’: an empirical investigation. Journal of Economic Psychology 45: 44–55.

48.

Stewart

(1983) Racial discrimination and occupational attainment in Britain. The Economic Journal 93(371): 521–541.

49.

Tomaney

Ward

(2000) England and the ‘new regionalism’. Regional Studies 34(5): 471–478.

50.

UK Census (2011) 2011 Census. Available at: https://www.ons.gov.uk/census/2011census

51.

Vanden Assem

Van Dolder

Thaler

(2012) Split or steal? Cooperative behavior when the stakes are large. Management Science 58: 2–20.

52.

Von Hippel

Silver

Lynch

(2000) Stereotyping against your will: the role of inhibitory ability in stereotyping and prejudice among the elderly. Personality and Social Psychology Bulletin 26: 523–532.

53.

Wall

(2011) Outwit, outplay, outcast? Sex discrimination in voting behaviour in the reality television show Survivor. New Zealand Economic Papers 45: 183–193.

54.

Whitaker

Lynch

(2011) Explaining support for the UK Independence Party at the 2009 European Parliament elections. Journal of Elections, Public Opinion & Parties 21: 359–379.

55.

Wikia (2017) Big Brother 1-17 (UK) and celebrity Big Brother 1-18. Available at: http://bigbrother.wikia.com/wiki/Big_Brother_UK_(franchise)

56.

Wikipedia (2017) Celebrity Big Brother 10. Available at: https://en.wikipedia.org/wiki/Celebrity_Big_Brother_10

57.

Wood

Hales

Purdon

, et al. (2009) A test for racial discrimination in recruitment practice in British cities. Research Report 607. Department for Work and Pensions. Available at: http://www.natcen.ac.uk/media/20541/test-for-racial-discrimination.pdf

58.

Wright

Ermisch

(1991) Gender discrimination in the British labour market: a reassessment. The Economic Journal 101: 508–522.

59.

Youtube (2017) Nominations footage. Available at: youtube.com/watch?v=vB0qCHVjYJ0&list=PL1017A958237238B9&index=78 (regular series 3, week 5); youtube.com/watch?v=LGkGWSESZv8 (regular series 6, week 10); youtube.com/watch?v=Kc9a1Tt5xO4&t=2529s (regular series 7, week 2); youtube.com/watch?v=D-kiwzEazLI&list=PL3d-fQNiJyhVlcT_hJx7Z6ym_57E7T3yG&index=96 (regular series 14, week 7); youtube.com/watch?v=4DdQPYo0RNw (regular series 16, week 3); youtube.com/watch?v=ZglHue_05n0&list=PL8E5BA57D8604ACD8&index=11 (celebrity series 3, week 2); youtube.com/watch?v=xN422gWM6ug&t=1775s (celebrity series 5, week 3); youtube.com/watch?v=hHDZ6jrklog (celebrity series 10, week 2); youtube.com/watch?v=MDthFAw_lp0&index=10&list=PLCG__CvNKQT3HNgka-AduMOSod7n-5Apq (celebrity series 17, week 1); youtube.com/watch?v=NjvW-OqN8lE (celebrity series 18, week 1)

60.

Zschirnt

Ruedin

(2016) Ethnic discrimination in hiring decisions: a meta-analysis of correspondence tests 1990–2015. Journal of Ethnic and Migration Studies 42: 1115–1134.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB

0.23 MB

5.19 MB