Hoop inequalities: Race,class and family structure background and the odds of playing in the National Basketball Association

Sampling frame

Our sampling frame comes from a complete listing of NBA players who were in the league and appeared on the official NBA website as of 15 July 2002. ¹ Of the total 475 NBA players listed, 465 had information that pertained to birth date and race. ² From this number, we selected those African American and white players who were born on or after 1977 in the United States and Canada (N = 155). We chose 1977 as the boundary for year of birth to focus our attention on a manageable subpopulation for whom we could –to the extent possible – control for selection bias based on newspaper reporting styles that may change over time. Given the birth year threshold, we found useful articles with sufficient biographical information between 1994 and 2004, a range that encompasses the sampled players’ experience from high school or college to when they were drafted and/or played their first few years in the NBA. We eliminated players who were born outside of the US and Canada because of possible country-level variations in the differences in, and effects of, class and family-structure background, which would be impossible to explain with the limited number of cases. The number of Canadians in our subpopulation is two. The racial composition of our subpopulation is roughly equivalent to the total NBA population in 2002: 86 percent of our sample are African American and 14 percent are white, compared with 77 percent African American and 21 percent white for the entire league. Overall, our percentages are substantively similar to those reported in other sources (e.g. Coakley, 2004), indicating that our method of identifying players’ racial background is reliable. ³

Coding

Classic sociological definitions of social class and family structure background guided our coding process of article text. To identify social class background, we examined words and phrases indicating parental occupation, description of neighborhood, and general statements of class position. Our definition of social class follows that of Weber (1946), referring to social groups whose position in society differ according to their relative access to economic resources. We define class primarily in economic terms (Myles and Turegun, 1994), relying heavily on parental occupation as the best single indicator of class background (Breiger, 1995). However, we used parental occupation only in cases where it was clear that the parents contributed to the player’s upbringing. Two examples illustrate journalistic phrases: ‘Father is an assistant principal … while his mother is a teacher’ and ‘[player’s mother] had worked her way up the corporate ladder at [company name] … Because of her success at [company name], [player’s mother] could transfer to just about anywhere in the country.’

Economic resources influence residential situation. Thus, we took neighborhood descriptions as a further marker of class position. ⁴ Neighborhood description consists of two possible indicators. The first is physical appearance of housing: ‘[Player] would run the 14 flights of stairs in his project’ or ‘… home in a two-car-garage corner of suburbia’. The second refers to perceived level and types of crime: ‘The dangers of the big city came to the [player’s childhood family] doorstep … a car drove past, shots were fired’, and ‘In the neighborhood we grew up in, you look around. Everybody is selling drugs.’

General statements of class position refer to players’ identification of relative economic standing, for example, ‘I was neither rich nor poor’ indicating, in this case, middle-class status. The other type was journalists’ direct reference to class position, for example, ‘he was raised in an upper middle class suburb’ to indicate above middle-class status.

Using information on parental occupation, neighborhood descriptions, and general statements of class position, our coding yielded three categories of social origin: 1) lower class, 2) middle class, and 3) upper middle and upper class.

To identify family background, we examined words and phrases indicating presence or absence of mother, father, or both. A non-grandparent two-parent family corresponds to two parents having raised the player. For this category, it did not matter if the father and mother were living together or are biologically related to the player. A necessary condition was that two parents were present and contributing during much of the respondent’s childhood upbringing. For most cases we found direct indicators (e.g. ‘raised by a single mother’), but some were indirect (e.g. ‘I never had a father around’). If player or reporter indicated ‘father was never around’, we assumed that the biological or step-father was either de jure or de facto separated from, and not contributing to, the household.

Considering the source of data, it was virtually impossible to determine the exact length of time a respondent was with or without two parents. Hence, we relied on media reports that clearly indicated who was present during a substantial portion of the pre-college or pre-NBA (depending on method of NBA entry) years. We treated step-parent families as two parents unless it was clear that the intervening single parent time spanned most of the respondent’s childhood. ⁵

We coded responses to family structure background as falling into four categories:1) raised by single father, 2) raised by single mother, 3) raised by grandparent(s) or social welfare institution, and 4) raised in non-grandparent two-parent family. ⁶

Difficulties in coding

In some cases indicators for social class and/or family background were blurry. Whenever descriptions did not fall clearly into a particular category, we did not code them, that is, we treated that player as missing data for that variable. The following description is an example of how information can yield no useful codes:

[player’s name] comes from the small town of [player’s hometown name], the second of eight children. His mother is on disability and his father is in … prison. [Player’s name] thinks often of his grandparents’ farm in [place name], where he spent his first seven years. He still likes to return in the summer, doing chores and fishing with his cousins.

Despite a rather detailed description of this player’s personal life and upbringing, parental occupation is not clear. One could assume that he belongs to the lower class, but cannot be sure: we do not know the respondent’s father occupation prior to imprisonment, nor the mother’s employment status prior to disability. Neither do we know the extent to which the grandparents provided support outside of summertime invitations to the farm. In addition, missing information on when the father went to jail prevented us from making a family-structure decision. As such, we treated this record for this player as missing data on both social class and family structure.

For more description of the matching of newspaper data to social class and family structure background, see the Appendix A on the Hoop Inequalities website.

Procedure to reduce missing data

Following Woolley (2000), word searches via Lexis-Nexis was a two-step process designed to reduce missing data. We first crafted words that were likely to be included in a player biography article: [player name], basketball, mother, mom, father, dad, grandparent(s), grandma, grandfather, parent(s), family, and class. We then refined the word search process based on our evaluation of the usefulness of the words and their combinations, noting that some words, like the player’s name and the word ‘family’, were more likely to yield social origin indicators. Variations on mother, such as ‘mom’, were used, providing some articles that we would not have found otherwise but oftentimes not yielding valid enough indicators of parental status. We used multiple word combinations in the search process for each player: for example, we used [player name], basketball, and mother in the same search, and in subsequent searches for that player, we replaced mother with father, parent, family, class and others we identified in the first step of the process.

Since not all newspaper articles provide the same type of information, we have different sample sizes for different variables. ‘Response rates’ for social class origin are 52 percent, and for family structure background 68 percent. For 46 percent of our respondents, we have information on both social origin variables.

Reliability checks are an important part of the process, and thus we used a common inter-rater reliability procedure. After gathering what appeared to be relevant articles, we each coded them independently. We then compared our codes, discussing similarities and discrepancies in light of our criteria for the coding categories. For social class, the inter-rater reliability percentage was 89, taking into account only those cases that had sufficient information with which to code. Similarly, our inter-rater reliability percentage for family structure background was 94. We resolved discrepantly coded cases through focused discussions. When discrepancies could not be resolved, the indicator was coded as missing. On average, we use two articles per player where social origin information could be coded; often, multiple indicators were found in the same article.

Newspaper based and indexing procedure selection bias

Selection bias stems from the contexts and practices of the news agencies in choosing which events are worthwhile reporting, and which are not. It creates ‘the difficulty … that one risks confounding the substantive phenomenon of interest with the selection process’ (Berk, 1983: 391). In the context of our data selection bias could stem from newspapers selectively reporting on a) some players biographies’ but not on others, and/or b) on some players’ social class and/or family structure background but not on others.

Whether the old media adage ‘if it bleeds, it leads’ is the mechanism, or the media’s tendency to present African Americans as poor (Gilens, 1996), there is reason to suspect that newspapers would be more likely to feature articles on players who faced extreme hardships, deliberately leaving out players who had very little newsworthy tragedy in their lives. In our data, almost all players had some aspect of their life history told in the newspaper, but variation in the length and depth of the story produces differences in the frequency of valid indicators of class and family structure.

To account for this possible problem, our total of 245 articles comes from the widest array of newspapers, ranging from local and regional (e.g. Sacremento Bee, Atlanta-Constitution Journal) to national (e.g. USA Today, New York Times). The form and magnitude of selection bias varies by newspaper constituency (Earl et al., 2004; Oliver and Meyers, 1999). We followed the argument that mixing newspaper types, even for the same player and especially at the local level, yields better data: ‘all things equal, the more local the focus of attention, the more credible the use of newspapers as a source of event data’ (McAdam and Su, 2002: 705).

Another form of newspaper selection bias occurs when the included cases are significantly different from NBA players for whom we do not have social origins information. We perform a common statistical procedure to address this issue: please see Appendix B: Assessing the impact of newspaper-based selection bias on the Hoop Inequalities website. We conclude that in terms of social class and/or family background, players for whom this information is available are not substantively different from players on which such information is lacking.

The practices of the online search index, in our case Lexis-Nexis, could also introduce selection bias, one potential problem being a reduction in the total number of possible available articles (see Woolley, 2000). ⁴ However, as Woolley points out, when used effectively, online indexing can be superior to consulting indexes created by newspaper publishers or private organizations (Woolley, 2000).

Description bias

Description bias refers to a situation in which the newspaper, for a variety of reasons including its own and/or its reporters’ prejudices, reports information incorrectly. Researchers of newspaper accounts of protest events note that description bias is less of a problem with hard news – that is, who, what, when, where, and why – where facts are more dutifully recorded and with less overt prejudice, and more of a problem with soft news – that is, editorializing of hard facts (Earl et al., 2004).

NBA players’ biographies can be either hard news or soft news, depending on the newspaper article author’s purposes. Players’ biographies have media value inasmuch as they are stories of ‘human interest and human drama’ (Oliver and Myers, 1999: 46). To accentuate newsworthiness, authors may provide their own judgments and embellish the story, thereby turning hard news into soft news; or, even worse, they may simply lie. With the type of data collection this article employs, it is impossible to determine the authors’ purposes. This is, perhaps, the greatest limitation of our methodological approach.

How serious is the threat of description bias for our analyses? We have no statistical means to detect incorrect information in newspaper authors’ depictions of NBA players’ biographies. What we do instead is look at the most likely sources of potential misrepresentation of players’ social background.

Trouble with reliability across indicators of social class and family structure background could occur if reporters apply sociological concepts ad verbatim, because they may use them in the wrong context. In our data, sociological terms rarely appeared as such: for example, out of the 245 analyzed articles, only five contained the term ‘class’ in reference to economic position or life chances. Usually, ‘class’ referred to descriptions of behavior (e.g. ‘class act’), or to school. In coding we were aware of potential misuse of sociological jargon, but noticed that little of it was present. Furthermore, our data collection approach relies on journalists’ strong suit, that is, detailed observations of the neighborhoods they visit and the people they meet, which are akin to hard news and can be successfully used in sociological work (see Earl et al., 2004 on this point).

Even if journalists are biased in the indicators that they report, the forms of bias are likely to be contradictory, in a sense canceling each other out. For example, research suggests that Americans tend to classify themselves as middle class when more objective measures indicate otherwise, and journalists may exhibit this behavior in their reporting (Jackman and Jackman, 1983; Luo and Brayfield, 1996). If this is so, our results would over-represent middle-class NBA players. On the other hand, given the media’s tendency to show African Americans in poverty, journalists may report more indicators of low class position, bending our results toward an overrepresentation of lower class African American NBA players (Gilens, 1996). Moreover, ‘rags-to-riches’ stories are popular American fare, compelling journalists to seek out these stories whenever possible (McManus, 1994; Shoemaker and Reese, 1996). If this is the case, then the number of NBA players who rose from poverty to NBA wealth and fame would be over-represented, rather than under-represented, in our data. While we cannot statistically examine description bias, each of the considerations above suggest that it does not substantially impact our findings.

Potential media bias – coupled with the fact that the media reproduces racial stereotypes – raises the issue of what, exactly, the units of analysis in these data are: are they real people, or representations of real people? In one sense, all data – whether it comes from in-depth interviews, content analysis or survey research – consist of representations influenced by the very procedures that brought the cases to light. All newspaper data – from political protest events to NBA players – are influenced by media processes. We followed procedures well-established in the large literature on newspaper data collection (e.g. Earl et al., 2004; Woolley, 2000) to systematically document and account for all possibilities of bias in our data collection procedure. These sources of bias do not appear to distort our results. In sum, we both captured detailed representations of real NBA athletes and feel as confident in our ability to generalize to the NBA population as social movement researchers feel in their ability to generalize from their representations of protest events.

Comparing NBA players to the US population: Sampling basedon characteristics of the dependent variable

Analyses in this article are performed on basketball players who have made it into the NBA. This raises the methodological issue of sampling based on the characteristics of the dependent variable, that is, as to whether we accurately account for the effects of the causal mechanisms on the dependent variable (Stolenzberg and Relles, 1997; Winship and Mare, 1992). Causality is defined here as a clear temporal chain of events where social origins (race, class and family structure) influence social destinations (attaining NBA status). Our causal model assumes an NBA sorting process as the mechanism for players’ selection into the league. Following sorting, which occurs before teams select players into the NBA, players are selected into the league (1) or not (0). In our approach, sampling on the dependent variable is akin to what Winship and Mare (1992) refer to as ‘truncated sample explicit selection’ (p. 330): we only have cases of the dependent variable selected into the league, and characteristics of those not selected are unknown.

To assess whether sampling on the dependent variable affects our substantive conclusions, we would need to know the characteristics of those who were eliminated by the NBA sorting process, and hence not selected into the league. According to our hypothesis, aspirants who come from advantaged social origins are more likely to be selected into the NBA than those who do not. If the majority of people not selected into the NBA were from lower social origins, then disadvantage decreases the odds of attaining an NBA position, in line with our expectation. If, however, most of those not selected were advantaged aspirants, then advantage decreases the odds of attainment, a situation contrary to our hypothesis.

We approximate the population of basketball aspirants as male children in the US We analyze two data sources containing these potential aspirants – the US Census 2000 and the General Social Survey (GSS) 1972–2002 – and compare the results with the newspaper data. ⁶

For the US Census 2000 data on male children, we define lower social class as a category at or below 150 percent of the poverty line in 2000. In 2000, the average poverty threshold was $17,608 for single person households and $26,412 for a family of four. Using these data we estimate that 45 percent of African American male children live in a lower social class context, while only 34 percent of African American NBA players come from lower social class background (z = −1.80, p < 0.05). ⁷ Similarly, 23 percent of white male children live in a lower social class context, while only 7 percent of white NBA players were from a lower social class background (z = −1.42, p < 0.10).

The differences in family structure between the relevant comparison group in the US population and NBA players are in the same direction, but less pronounced. Among African American male children, 62 percent come from non-two parent families, while in the NBA sample, 57 percent of African Americans do (z = −0.95, p = 0.17). ⁸ Similarly, 26 percent of all white male children and 19 percent of white NBA players come from non-two parent families (z = −0.63, p = 0.26).

For compounded disadvantage, we used the General Social Survey 1972 to 2002, including only males. For this data set, we defined lower social class as the category of respondents reporting below-average family income level at age 16. For family structure background, we created a dummy variable distinguishing two-parent family from the rest.

For African Americans, 28 percent of the NBA players come from compounded disadvantaged backgrounds, compared to 39 percent in the GSS sample (z = −1.66, p < 0.05, one-tail test). None of the white players in our study come from compounded disadvantaged origins, compared to 30 percent in the GSS sample (z = −2.17, p < 0.01). These results indicate that NBA players as a whole are less likely to come from compounded disadvantaged backgrounds than their counterparts from the general population.

With regard to social class, family background, and the combination of the two, comparisons between potential basketball aspirants and NBA players indicate that the latter are significantly better off in terms of social origins than the former. This finding is in line with our research hypothesis that the advantaged are more likely to be selected into the league than the disadvantaged, and suggests that in this study sampling based on characteristics of the dependent variable did not substantially alter the substantive conclusions.

To further illuminate how the intersection of race, class and family structure background influences the chances of NBA attainment, we used data from the US Census 2000 and the General Social Survey to calculate odds ratios. In terms of odds ratios, we estimate that among African Americans, a child raised in lower class family has a decreased likelihood of 37 percent to become an NBA player than a child raised in middle or upper class family. An analogous figure for whites shows that children from lower class families are even more disadvantaged (decreased likelihood of 75 percent). In the case of non-two-parent families, the coefficients for African American and whites are more similar than in the case of social class. Among African Americans a child from a non-two-parent family has 18 percent lower chance to become an NBA player than a child raised in two-parent family. For whites the figure is 33 percent.