Who Creams? Explaining the Classroom Cream-Skimming Behavior of School Teachers from a Street-Level Bureaucracy Perspective

ABSTRACT:

Ideas as to how and why individuals resort to creaming are generated primarily by a few qualitative studies and have, to our knowledge, not been tested quantitatively. This article aims to fill this gap and explains the classroom cream-skimming behavior of school teachers in Denmark, defined as prioritizing the teaching of academically promising students. Drawing on the street-level bureaucracy literature, it tests the following propositions: (1) creaming is directly related to an inadequacy of resources, and this relationship is moderated by the breadth of parental involvement in their children’s education; (2) creaming is weakly related to the presence of bureaucratic success criteria; and (3) creaming is directly related to the level of parental involvement in and satisfaction with their children’s education. These are tested using data from a 2011 survey of Danish and mathematics teachers in public and private schools across Denmark, and a class-fixed-effects design. Overall, the findings provide varying support for these propositions.

NOTES

ACKNOWLEDGMENTS

The author would like to thank Hans Henrik Sievertsen, Alessandro Martinello, and members of the project “School Management, Teaching and Student Performance” led by Søren C. Winter, especially Larry O’Toole, Simon Calmar Andersen, and Mogens Jin Pedersen, for their help in improving this article.

Notes

1 However, there is considerable qualitative literature on the phenomenon, albeit using a different terminology. Studies in education have documented, via classroom observation, teachers’ differential treatment of students through their verbal and non-verbal behavior, both at the individual and group levels and often in subtle ways (see, e.g., Babad 1993; Babad and Taylor 1992).

2 In the updated version of his book (2010), Lipsky slightly changed his notion of who is (and who is not) an SLB. He notes that “not every teacher … experiences the pressures that I stated street-level bureaucrats face by definition. Frontline workers whose jobs are relatively free of restrictive structural constraints will still develop routines in response to their work environments. But the routines will not be developed to with a difficult work environment” (Lipsky 2010: xvii). Lipsky thereby implicitly acknowledges that a chronic shortage of resources is not a defining criteria of an SLB and that the routines SLBs employ can also be influenced by other factors, such as attitudes.

3 In their recent conceptualization of “coping” based on a systematic review of the literature on the concept, Tummers et al. (2015), drawing on the classification system of Skinner and her colleagues, identify three distinct families of coping: moving towards, away from, and against clients, respectively. They classify creaming (which they refer to as “prioritizing”) as part of the coping family “Moving towards clients” and in the sub-category “Prioritizing among clients,” which they describe as “Giving certain clients more time, resources, or energy” (Tummers et al. 2015). (They use the term "prioritizing" instead of “creaming” on the grounds that the latter term has a negative connotation.) However, prioritizing is not only about moving towards clients; it is also about moving away from clients. Prioritizing, by definition, means giving one thing precedence over another. Following this line of reasoning, creaming is a form of rationing, which is defined as “decreasing service availability, attractiveness, or expectations to clients or client groups,” and thus also belongs to the coping family “Moving away from clients.”

4 Definitions of creaming can vary, but their essence remains the same: the selection of clients on the basis of given performance standards and the accompanying incentives (either explicitly provided or arising naturally from these). For example, Heinrich (1999) summarizes a number of studies that found that incentives generated by cost-per-placement standards in employment agencies have led to increased provision of short-term, less intensive services, an emphasis on immediate placements, and the selection of more job-ready participants who require less training. Mukamel et al. (2009) find some evidence that nursing homes responded to the publication of a national quality report card for nursing homes by adopting cream-skimming admission policies; similarly, Berta et al. (2010) examine cream skimming in response to a prospective payment system in the health care sector in Italy and find that private hospitals are involved in cream skimming at a much higher rate than public and not-for-profit hospitals.

5 Information from Ministry of Education: http://uvm.dk/Administration/Tilskud-til-institutioner/Fakta-om-taxametersystemet-og-tilskud-til-institutionerne.

6 Such behavior may also be facilitated by the Danish scale of assessment: –2, 0, 2, 4, 7, 10, 12, and where 0 or below means failed. In the conditions described ear;oer, a teacher may, for example, choose to focus on improving the performance of those scoring around 7 or 10 at the cost of those scoring 0 or –2 because of the lower bound. There is more to be gained in improving the performance of 7-scorers (e.g., a three-point increase up to 10) than to prevent those scoring 2 or 0 to slipping a level (a two-point drop).

7 The specifics of the performance measure might make a difference in some systems. For example, in testing systems in the US, the metric often employed is overall pass rate for a classroom (or grade), rather than average score on the exam. The latter might encourage more creaming than the former, or the former might even encourage reverse creaming to get several marginal students above the threshold.

8 Scholars generally recommend listwise deletion instead of pairwise deletion. As Allison (2009) explains, pairwise deletion allows you to use more of your data. However, each computed statistic may be based on a different subset of cases. This can be problematic. This procedure is sensible if (and only if) the data are randomly missing. In this case, each correlation, mean, and standard deviation is an unbiased estimate of the corresponding population parameter. This is very rarely the case. If data are not missing at random (the typical scenario), several problems can develop: The pieces put together for the regression analysis refer to systematically different subsets of the population; e.g., the cases used in computing the correlation between variables 1 and 2 (r12) may be very different than the cases used in computing r34. As a consequence, results cannot be interpreted coherently for the entire population or even some discernible subpopulation. The more common problem, however, is the difficulty in getting accurate estimates of the standard errors. That is because each covariance (or correlation) may be based on a different sample size, depending on the missing-data pattern. Listwise deletion, in contrast, has the disadvantage that it often discards a great deal of potentially usable data. On the one hand, this loss of data leads to larger standard errors, wider confidence intervals, and a loss of power in testing hypotheses. On the other hand, the estimated standard errors produced by listwise deletion are usually accurate estimates of the true standard errors. In this sense, listwise deletion is an “honest” method for handling missing data, unlike some other conventional methods. For these reasons, listwise deletion is generally recommended instead of pairwise deletion.

9 The internal consistency of the variable as measured by Cronbach’s alpha was 0.60, 0.57, and 0.60 for the full sample, the public schools sample, and the private schools sample, respectively.

10 The internal consistency of the variable as measured by Cronbach’s alpha was 0.75, 0.74, and 0.75 for the full sample, the public schools sample, and the private schools sample, respectively.

11 The eight items loaded on two separate factors: compassion and public service commitment. Including these separately in the model did not make a substantive difference to the results. Therefore, they were included as one overall indicator of commitment to public service.

12 The internal consistency of the variable as measured by Cronbach’s alpha was 0.79, 0.80, and 0.77 for the full sample, the public schools sample, and the private schools sample, respectively.

13 Attempts to estimate the models using ordered logit with fixed effects failed as the models failed to converge.

14 In addition to the analysis underlying the results in Table 1, we ran several models with different specifications in order to test a range of interaction effects; e.g., whether the relationship between resource shortage and creaming was moderated by teachers’ assessment of their relationship with parents or by school management’s expectations regarding students’ academic performance. None of the tested interaction effects was significant.

15 One way of indirectly assessing the validity of these measures is to see how they correlate with related measures from other sources. The measure of time constraints correlates .13 (p < .05) with a measure of the socioeconomic composition of the class generated from administrative data, indicating that the lower the proportion of the class ranked among the top 50% of all students in the dataset with respect to parents’ SES, the greater the creaming. We can also validate the measure of academic expectations by comparing the responses of teachers teaching the same class to the same question. Thirty percent of the responses match and, of those that are different, more than 75% have only a one-unit difference on the scale; in addition, the teachers’ responses on one of the two items (performance in exams compared to other schools) correlate well to that of their principal (0.32, p < .001).

16 This is calculated as follows: Range of variable teaching time shortage = 16.01 (from –0.46 to 15.55), unstandardized coefficient (b) = 0.185. Therefore, effect of the variable across its full range in the sample (i.e., as it moves from its minimum value to its maximum value) = 16.01 x 0.185 = 2.96. Range of variable parental involvement = 3.85 (from –2.75 to 1.1), unstandardized coefficient (b) = 0.242. Therefore, effect of the variable across its full range in the sample = 3.85 x 0.242 = 0.932.

17 In our data, the (teacher-reported) percentage of engaged parents in public and private schools is, on average, 69.82 (std. deviation: 25.89, range: 1–100) and 79.40 (std. deviation: 23.59, range: 15–100), respectively, a difference of almost 10 percentage points.

18 In these conditions, teachers’ public service motivation in the form of a focus on the needs of all students may have a greater role to play as a counteracting force. As noted earlier, the PSM coefficient for private school teachers, though insignificant, has the correct sign and substantial magnitude. This result deserves further investigation in future studies.

19 First, the survey items used to construct the dependent and main explanatory variables are spaced out across the survey, worded very differently and with different response choices, and cannot be examined simultaneously, given web interface. In other words, the item measuring creaming behavior (the dependent variable) is question 22 in the survey (response choice on a scale from Strongly disagree to Strongly agree), the index of infrastructural resource shortage (H1) is constructed from question 9 (response choices from Not at all to To a very large extent), and the measures of parental involvement and satisfaction (H3) are questions 35 and 36. Parental involvement is indicated by a single number, chosen by the respondent to indicate the percentage of parents in the teacher’s class who take an interest in their child’s schooling. Parental satisfaction is one item in a battery of four (the first three of which deal with students’ academic level and well-being) and with response choices from Much below average to Much above average. Teachers’ beliefs about school management’s expectations regarding academic standards (H2) are formed by summing two items worded differently and spaced widely apart in the survey and with different question formats: one in the form of two statements anchoring opposite ends of a five-point scale and the other in the form of a single statement with a five-point response scale going from Strongly disagree to Strongly agree. Thus, there is no similarity between wordings of dependent and independent variables that should give great cause for concern. In addition, we have, wherever possible, chosen measures based on questions that elicit concrete information from the respondents to increase reliability and validity. In other words, the variable measuring teaching-time shortage is constructed from a battery of four items that ask the respondent to provide more precise information in the form of an estimate of the proportion of lesson time used on different activities, such as maintenance of order in the class, teaching, and administrative practical tasks. We have intentionally avoided asking teachers directly about their workload; e.g., in the form “On a scale of 1–5, where 5 = very high, how high do you think your workload is?” Or in the form “I do not have enough time to finish my duties,” please respond on a scale from 1–5, where 1 = strongly disagree to 5 = strongly agree. Response set issues should be minimal, given that the survey questions do not involve sensitive topics such as violent or aggressive behavior, substance abuse, or sexual practices. Further, the survey was administered by mail by a well-regarded research institution, which ensured anonymity and increased the respondents’ trust regarding the purpose of the data collection. An examination of responses to individual items did not reveal any patterns suggesting a response set.

20 Teaching-time shortage is a significant predictor of creaming, regardless of the other variables included in the model. In addition, in both the all-schools sample and the public schools sample, the magnitude and significance of the variable Percentage of Parents Involved in their Child’s Schooling increases slightly after all of the standard controls, such as teacher’s sex, education, and experience, are included. The same applies in the case of the variable Parental Satisfaction with Schooling in the private schools sample.

Additional information

Notes on contributors

Siddhartha Baviskar

Siddhartha Baviskar ([email protected]) is a lecturer at the Institute of Social Work, University College Copenhagen. His current research focuses on the design of social interventions for the benefit of children placed in foster care and their families.