Peer Review Articles on Purpose of Schooling Standardized Testing

• Periodical List
• Front Sociol
• five.5; 2020
• PMC8022478

Front Sociol. 2020; v: 544628.

Standardized Testing, Use of Assessment Data, and Depression Reading Operation of Immigrant and Non-immigrant Students in OECD Countries

Janna Teltemann

¹Section of Social Sciences, University of Hildesheim, Hildesheim, Germany

Reinhard Schunck

²School of Human being and Social Sciences, University of Wuppertal, Wuppertal, Federal republic of germany

Received 2020 Mar 21; Accustomed 2020 Oct 23.

Abstract

This paper investigates the furnishings of standardized testing and publication of achievement data on depression reading functioning for immigrant and not-immigrant students in xxx OECD countries. The paper aims to test hypotheses derived from a principal-agent framework. According to this theoretical perspective, standardized assessments lone should not be associated with reading performance. Instead, the model proposes that the provision of the results to the principle (parents and pedagogy regime) is associated with higher student performance, as this reduces the information asymmetry between principal (parents and educational authorities) and amanuensis (teachers and schools). The results of our analyses of PISA 2009 and 2015 reading data from 422.172 students bear witness that first, the use of standardized accomplishment tests lonely was not associated with the risk of low performance. 2nd, making the results of standardized tests bachelor to the public was associated with a decreased adventure of low reading performance amid all students, and, 3rd, peculiarly among first generation immigrant students. These results were robust across various modeling approaches. In accordance with the predictions from the principal-agent framework, our findings advise that the mere implementation of standardized assessments has no effects on depression performance. Testing forth with the public provision of the testing results, which decreases the data asymmetry between schools and teachers on the i hand and parents and education authorities on the other, was associated with a decreased risk of low performance, with the effect being stronger for immigrant students.

Keywords: immigration, didactics, standardization, PISA, educational inequality, principal-agent model, stock-still furnishings, longitudinal analyses

Introduction

Integrating growing immigrant populations is a claiming for receiving countries. Since pedagogy is a key resource in gimmicky societies information technology is also a central to societal integration of immigrants and, in item, their descendants. International large scale assessments such as the OECD PISA study take drawn attending to countries' educational activity systems and how they may contribute to educational inequalities and differences in integration processes. As pressure level for quality and equity in education increased, policy making in instruction has been under shut monitoring during the last years. A major focus of educational reform in many countries has been the implementation of educational standards and, in detail, their regular assessment through nationwide standardized testing (Scheerens, 2007; Meyer and Benavot, 2013). Standardized testing is supposed to aid the definition of clear educational goals and serves as a measure of accountability (i.e., the enforcement of responsibilities to attain these goals), which, in plow, are believed to affect incentives, restrictions, and opportunities of the actors involved in "producing" education. This rationale is fatigued, in function, from principal-agent-models which are based in rational pick theories of individual activeness (Wößmann, 2005; Levačić, 2009). While master-agent-models are frequently referred to in empirical research using large scale assessment data like PISA, their mechanisms are rarely put to a direct test. More ofttimes, these models are mentioned in order to explicate a possible empirical association between standardized testing and educational outcomes.

In this newspaper, we add to the literature by, outset, testing mechanisms drawn from a main-agent model of education more than directly. To do and then, nosotros investigate if the use of nationwide standardized testing affects pupil functioning, and, more importantly, if reporting the results of such assessments to the public or educational authorities does. From the perspective of principal-amanuensis models, we would expect that reporting of the results is particular important, since it reduces the data disproportion between the agent (schools and teachers) and the principal (parents and educational authorities). Second, we take a closer look at immigrant students. The number of immigrants has increased substantially in most Western receiving countries during the last years. Third, because we focus on immigrant student, nosotros do non examine average achievement as an upshot but the take chances of low reading performance. This is divers as functioning below the second proficiency level in reading in PISA. Reaching this level of reading proficiency is necessary to participate effectively in lodge and tin thus exist seen as a prerequisite for immigrant integration. Not reaching this level of proficiency is related to lower life chances: Follow-up studies based on PISA have shown that performance below this level is related to a lower chance of transition to post-secondary instruction and a higher take a chance of unemployment and income poverty (OECD, 2010; Shipley and Gluzynski, 2011). Fourth, we use a longitudinal design at the country level by using data from the OECD Programme for International Student Assessment (PISA) 2009 and 2015 from 30 OECD countries. The longitudinal blueprint allows u.s. to control for (time constant) unobserved land characteristics, making the estimates less prone to bias.

The remainder of this paper is structured as follows: in the next section, we elaborate our theoretical arguments on the furnishings of standardized testing based on the principal-agent model. Thereafter, we summarize findings from previous studies on the impact of testing practices on educational outcomes. In section "Data and Methods" we draw our database and methods. After presenting the results in department "Results", nosotros discuss implications and limitations of our report in the final section.

How Standardized Testing Can Touch on Performance—Theory and Hypotheses

From a rational choice perspective, institutions of the educational activity system affect incentives, restrictions, and opportunities of the actors involved (i.e., students, parents, teachers, principals). Following this rationale, education policies aiming for quality education should exist most effective if they take implemented institutional regulations which incentivize high attempt of the actors involved (eastward.k., teachers). Rational pick models of education further presume that actors, in our case teachers, may not necessarily exist interested in high performance and may aim to avert extensive effort. Parents and the state, even so, wait schools and teachers to invest try in teaching in order to realize quality education. This is a classic master-agent constellation (Laffont and Martimort, 2002): A principal, the parents and/or the administrative authorities, commissions an agent, the school, to do something on their behalf, i.e., to provide instruction to the students (Ferris, 1992; Wößmann, 2005; Levačić, 2009).

The principal-amanuensis framework draws attention to three possible problems (Jensen and Meckling, 1976): First, the agents' and principals' preferences may not align. Second, there is an asymmetry in information—oftentimes the principal cannot observe the agent'due south behavior directly. Tertiary, the chief has to be able to evaluate the agent's behavior, i.e., he needs to appraise how much effort the agent puts into realizing the primary'southward goals. Therefore, for master-agent constellations to work in the master's interest, at to the lowest degree two weather have to be met. Get-go, the principal's goals have to be conspicuously defined in order to be realized. This is 1 of the justifications for the specification of national standards in education. They are supposed to clarify the goals of education and office as a frame of reference and orientation for the actors involved (Klieme et al., 2003). Second, it is not sufficient to just spell out the educational standards, they also need regular assessment. Hence, a ofttimes used indicator of the standardization of an didactics system is the use of regular (nation-wide) standardized tests.

A main argument in the literature is that standardized tests improve overall functioning (Wößmann et al., 2009; Bol et al., 2014). The theoretical mechanisms governing this effect are even so oftentimes rather implicit; mostly, it is assumed that the mere existence of such tests tin can either cause a grade of "gentle pressure" on schools and teachers and their manner of teaching or increase the signaling value of educational credentials (for a notable exception and an explicit theoretical model, meet Bishop, 1995)¹. Information technology is argued, for case, that if teachers practise non know which tasks are assessed in tests—because the tests are conceptualized past a central authority—they will be less probable to skip parts of the curriculum and the content taught will be more comprehensive (Wößmann et al., 2007, p. 25f.).

However, from a theoretical point of view, this mechanism appears incomplete. The implementation of standardized testing itself is non sufficient to resolve the chief-agent problem, as information technology does not affect the data asymmetry betwixt both parties. The main needs to have information on the results of the standardized tests. The more information the principal has, east.g., achievement information of other schools or national averages, the improve volition the principal exist able to evaluate the agent's beliefs and sanction information technology, positively or negatively. Thus, simply if the results of the standardized tests are available to the principal, will there be a relevant decrease in the asymmetric relation. From the logic of the principal-amanuensis model, this form of accountability increases the agents' incentives to human activity according to the principals' preferences. Consequently, schools and teachers every bit agents are confronted with a higher pressure to improve their students' accomplishment. We therefore expect lower rates of low performing students in countries where assessment results are communicated to the public or administrative authorities (Hypothesis 1).

Furthermore, when it comes to the risk of low performance, different students take different risks. Immigrant students, for instance, are oftentimes in need of special private (language) support. As their parents have less knowledge about the rules of the education arrangement, teachers, and schools accept to invest more fourth dimension for consultation. The specific situation of immigrant students creates a higher demand for teachers and, from the perspective of the principal-amanuensis model, a college risk for opportunistic behavior (east.g., negligence of the specific needs of immigrant students). If, however, accomplishment data is available to the principals, this creates stronger incentives for schools to take care of every pupil, regardless of their groundwork. The existence and publication of the results of standardized tests therefore should be advantageous for immigrants.

Further, we argue that information technology is rational for schools to concentrate efforts on those student groups who are in detail need for help (such every bit immigrant students) (Motiejunaite et al., 2014), as their performance may have a potent affect on a school's mean operation level. Findings from research on the effects of standardized assessments in the USA showed that for some tests and tasks, adaption of teaching strategies was more prevalent in schools with larger shares of ethnic minorities and low performing students (Mittleman and Jennings, 2018). Further, in some countries, standardized assessments are targeted toward minimum levels of educational activity. As a consequence, teachers may particularly focus on students who are at risk of not reaching this level (Booher-Jennings, 2005), which ofttimes are immigrant or indigenous minority students. In the context of low educational performance, we thus expect immigrant students to turn a profit more from standardized testing and a publication of cess results than not-immigrant students (Hypothesis ii).

Furnishings of Standardized Testing on Achievement—Previous Results

Since the publication of the starting time PISA round in 2000, a number of studies investigated how aspects of educational standardization are related to pupil accomplishment and inequality in educatee accomplishment (Schütz et al., 2007; Horn, 2009; Chmielewski and Reardon, 2016; Bodovski et al., 2017). These studies often focused on standardized testing, which is seen as 1 aspect of an education system'south degree of standardization (Bol and van de Werfhorst, 2013). It has to exist noted, still, that standardized testing should non be used alone to evaluate the caste of standardization of a country's instruction system. To assess if an education arrangement can be described every bit standardized, other dimensions of (de)standardization, such as curriculum standardization, school autonomy (in selecting teachers, allocating resource, etc.), and the modes of instructor didactics, have to exist considered as well. Since our focus lies on standardized testing—and not standardization in general—we concentrate the literature review on studies that either focus on this dimension or on immigrant students.

Several previous studies looked at the consequence of primal schoolhouse exit exams, which are a special type of a standardized assessment, and mostly found that they are associated with higher average test scores (Bishop, 1997; Carnoy and Loeb, 2002; Wößmann, 2003; Fuchs and Wößmann, 2007). Bergbauer et al. (2018) compared the furnishings of standardized external comparisons and standardized monitoring to effects of more internal developed testing procedures, using data from six unlike PISA studies (2000–2015). Their results show that standardized external comparisons as well equally standardized monitoring are associated with higher levels of competence among students. Cartoon on data from TIMSS 1995, Jürges et al. (2005) analyzed the effect of primal exit exams on achievement scores in lower secondary teaching in Frg. They establish that students in federal states with central exit examinations outperform students in states without central schoolhouse leaving assessments.

A small number of studies addresses the effects of testing on immigrant accomplishment and, to the best of our cognition, at that place are no existing studies that focus on assessments and on the educational inclusion of immigrant students in terms of functioning below a certain threshold. Schneeweis (2011) found significant (positive) effects of external educatee assessments on immigrants educational accomplishment only for OECD countries. Cobb-Clark et al. (2012) found insignificant furnishings of external examinations on examination score gaps betwixt immigrants and natives, but for ane of eight assessed groups they estimated a meaning negative event. Teltemann (2015) constitute smaller achievement gaps in countries where accountability measures were implemented. Wößmann (2005) reported positive effects of central exams for low achieving students, suggesting that central exams bring an advantage for immigrant student and students from less-educated backgrounds.

Data and Methods

We draw on data from the 2009 and 2015 OECD Program for International Student Assessment (PISA, OECD, 2016). Both PISA rounds comprise information on testing procedures and the publication of the testing results. Since its kickoff survey in 2000, PISA is the nigh regular and wide-ranging competence assessment of secondary school students. In 2015, more than 540,000 students in 72 countries have been tested. PISA assesses curriculum-contained competences in reading, mathematics and science. In addition, PISA collects a broad range of groundwork information by administering context questionnaires to students, parents, and principals. The sampling blueprint is targeted at a representative sample of the 15-years erstwhile school population in a country, contained of the respective grade they are attending. PISA is conducted every 3 years and the PISA datasets are publicly available via download from the OECD'due south websiteⁱⁱ. Since we pooled the information from 2009 and 2015, we created a data construction with four levels: students, schools, land-years, and countries (encounter the section on Modeling beneath). All analyses were carried out using Stata 16.1. Code for reproducing the analysis have been archived on the Open Scientific discipline Framework (https://osf.io/3ezxs/).

Dependent Variable

With regard to immigrant integration, the definition of competences in PISA, which does not target national curricula merely seeks to measure "viability" in globalized economies, proves useful. The PISA competence scores "measure out how far students approaching the end of compulsory education accept caused some of the cognition and skills essential for full participation in the knowledge society" (OECD, 2009b, p. 12). Thus, assessing differences between immigrant and non-immigrant students with PISA data can give insight not merely into educational integration but likewise into future societal integration. Competences in PISA are measured on a continuous scale which is standardized to an OECD mean of 500 points. In addition, PISA distinguishes and then-called proficiency levels, which correspond to actual abilities. For reading, proficiency level 2 is divers as a baseline level of competences, "at which students begin to demonstrate the reading skills that will enable them to participate effectively and productively in life" (OECD, 2016, p. 164). Functioning below this baseline level thus indicates the risk of failed societal integration for immigrant students, as has been shown past PISA follow-up studies (OECD, 2010; Shipley and Gluzynski, 2011). PISA provides several (five up to 2012 and ten since 2015) plausible competence scores per student (come across OECD, 2009a for details). We used the (first) 5 plausible values and created dummy variables that signal operation below proficiency level 2 (a score below 408 points, run across OECD, 2009a, p. 117ff.). Consequently, the final coefficients represent the boilerplate over five models (Macdonald, 2019).

Primary Independent Variable and Controls at the Student Level

In PISA, immigrant status is assigned according to the country of birth of a educatee and its parents. Students who indicated that they and their parents were built-in away are categorized as first generation students. Second-generation students were born in the land of exam with both parents born abroad. Since PISA does not collect comparable or complete information on students' or parents' countries of origin—the manner this is inquired differs between the participating countries—we cannot distinguish different immigrant groups. This is a major drawback, since the composition of immigrant groups may covary with the receiving countries' contextual conditions, including their educational institutions. To alleviate this problem partially, we control for linguistic communication apply at home with a dummy variable indicating whether students reported to mainly speak a foreign linguistic communication and not the test language at home. Furthermore, considering migration into OECD countries may exist selective on socioeconomic status, we also control for several measures of parental socioeconomic groundwork. This includes parental education (measured through the ISCED calibration), family wealth possessions (measured through the "wealth" index in PISA), cultural possessions (measured through the "cultposs" index in PISA), and home educational resource (measured through the "hedres" index in PISA) (see OECD, 2017, p. 339 for details). Lastly, we control for educatee gender (ane = female person).

Main Independent Variables and Controls at the Country-Year Level

Following the arroyo described by the OECD (OECD, 2013, p. 28, 66, 166), nosotros have aggregated school data inside countries for 2009 and 2015 to describe the system level. This is possible since PISA draws a representative sample of schools and the schools' principals accept been interviewed about organizational aspects of their school. For each year we constructed three variables according to this procedure: commencement, the proportion of students in a country attending schools that regularly administer mandatory standardized tests. Second, the proportion of students attention schools that post aggregated achievement data publicly and, tertiary, provide aggregated achievement data to educational regime³.

A country's institutional arrangements are non independent of other country characteristics that might also affect student achievement. Since nosotros are employing a longitudinal approach at the land level and include land fixed effects (see Modeling below), all time-constant country differences are accounted for. Even so, outcome estimates may still be biased by time-varying differences betwixt countries that covary with standardized testing and pupil performance. We therefore control country characteristics that may simultaneously affect (immigrant) student performance and are related to the country'south institutional arrangements. In lodge to control for a full general effect of resources devoted to the educational organisation, nosotros include annual educational expenditure as a per centum of a land's Gross National Income in our models. Likewise, nosotros control for effects of economic development of a country by including the annual growth of a land'southward Gdp (in percent). The overall number of immigrants in a state may be related to institutions, such as integration policies, which might accept an impact on educational performance of immigrants. We therefore control for the international migrant stock equally a percentage of the overall population. Additionally, immigrant functioning may be impacted by their labor market outlooks. Hence, we control for the annual unemployment rate among foreign built-in persons in each land. Data for these almanac country-year control variables comes from the World Bank and the OECD (Fontenay, 2018). An overview on the distribution of these characteristics amidst the countries in our sample can exist found in the Appendix (Table A1) also equally their pairwise correlations (Table A3).

Analyses Sample

We restricted our analyses to OECD countries in order to increment comparability across countries. We excluded countries for which (state-level) information was unavailable and those with <40 immigrant students (either commencement or 2nd generation) in the sample—this practical to Japan, Korea, Poland, and Turkey. Students were excluded if they had missing values on any variable (listwise deletion). Our final sample consists of 422.172 students in 12.255 schools in 54 country-years in 30 countries. Table i gives an overview over unweighted sample statistics.

Table 1

Sample statistics (unweighted).

	Hateful	Sd	Min	Max
Student level variables
Below reading level ii (pv1)	0.18		0.00	1.00
Below reading level 2 (pv2)	0.nineteen		0.00	1.00
Below reading level 2 (pv3)	0.18		0.00	1.00
Below reading level 2 (pv4)	0.18		0.00	1.00
Beneath reading level 2 (pv5)	0.xviii		0.00	1.00
Native	0.89		0.00	one.00
First generation	0.05		0.00	ane.00
Second generation	0.06		0.00	1.00
Gender [ane = female]	0.51		0.00	1.00
Linguistic communication of test spoken at home	0.88		0.00	ane.00
Parental instruction
None	0.01		0.00	1.00
ISCED 1	0.03		0.00	1.00
ISCED 2	0.10		0.00	1.00
ISCED 3b,c	0.08		0.00	ane.00
ISCED 3a,iv	0.24		0.00	1.00
ISCED 5b	0.17		0.00	1.00
ISCED 5a,6	0.37		0.00	ane.00
Index of family wealth possessions	−0.01	1.05	−7.44	four.44
Index of cultural possessions	−0.02	0.98	−1.92	2.63
Alphabetize of home educational resources	−0.05	i.00	−iv.45	1.99
Country level variables (source WB)
International migrant stock (% of population)	12.79	8.14	0.82	43.96
Adapted savings: education expenditure (% of GNI)	v.03	0.93	3.10	8.34
Gross domestic product growth (annual, %)	−1.23	4.74	−fourteen.43	25.16
Unemployment (%) among foreign built-in	11.63	6.eighteen	4.xxx	32.00
Proportion of students attending schools that (PISA aggr.)
Regularly use mandatory stand up. tests	0.73	0.21	0.24	ane.00
Post accomplishment data publicly	0.39	0.24	0.02	0.92
Provide adm. authority with achievement information	0.69	0.22	0.26	0.99
PISA round
PISA 2009	0.58
PISA 2015	0.42
North	422,172

Source: PISA 2009, 2015, Earth Banking concern.

Modeling

As our dependent variable is binary and our data construction is clustered hierarchically, we estimated 4 level linear probability models (LPM). The individual students (level i) are clustered in schools (level 2), which are clustered in country-years (triennial land observations) (level iii), which are once more clustered in countries (level 4). The standard approach to this data structure is a four-level random furnishings model

y i j g 50 = β 0 + β 1 x i j thou l + β 2 c thousand 50 + β three x i j k 50 × c k 50 + t + w fifty + v chiliad 50 + u j k fifty + ε i j k fifty

(ane)

where the dependent variable y _ijkl is the probability of an individual student i in school j in country-yr k in country l to fall below PISA reading level ii. due west _l represents the country-level error, v _kl the land-twelvemonth error, u _jkl the school, and ε _ijkl the educatee-level error. x _ijkl exemplifies the private-level variables (i.e., migration groundwork, gender, language power, and parental socio-economic status) and t represents joint period (wave) effects. The effects of involvement are those associated with the state-year–specific variables (β₂) and their interaction with immigration status (β_iii).

Although we focus on OECD countries, the country sample is still heterogenous with respect to clearing histories, institutional arrangements, educational policies, and economic conditions, all of which may be correlated with aspects of the education system and, in particular, testing and accountability. Thus, the problem of unobserved heterogeneity at the country level is pressing and the probability of misspecifying the model is high. The standard strategy to avoid misspecification is to command for the relevant confounders. However, the power to include relevant confounders is restricted for ii reasons. Offset, with 30 countries (and 54 country-years), the degrees of freedom are limited. Second, many important confounders, due east.g., which describe a country'south immigration history, are not readily measured and available. Therefore, we estimated (1) as a first departure (i.eastward., stock-still effects) model (Wooldridge, 2010), including fixed effects for countries and years. The advantage of the fixed effects approach is that we do not have to make any assumptions nearly possible confounders at the country level. The model thus produces unbiased estimates fifty-fifty if there are unobserved confounders at the country level—that is, E(w _fifty |x _ijkl , c _kl ) ≠ 0. Therefore, the effects of the country-twelvemonth level variables are estimated solely by relying on within-country (co)variation.

The coefficients in the LPM are estimators of the absolute difference in the probability of low reading achievement associated with a unit increase in the value of the corresponding predictor variable. We have chosen a linear probability model over a logistic model for the following reasons. First, the available non-linear four level models in the statistical program used for the analyses (Stata) do not take weights. Weighting the information, however, is necessary in view of the complex and nationally diverging sampling procedures in PISA (OECD, 2009a; Lopez-Agudo et al., 2017). Second, non-linear models are notoriously hard to interpret, in particular when dealing with interactions. One needs to estimate average marginal effects in order to empathize the joint outcome of main- and interaction outcome (Brambor et al., 2006; Drupe et al., 2012). While other statistical software packages (east.one thousand., MLwiN) are able to guess weighted four level logit models, they are unable to provide boilerplate marginal furnishings. Third, an important argument against the LPM is that it may provide predicted probabilities >i or <0 (Long, 1997). Withal, in many situations, the LPM is applicable (Hellevik, 2009) and, as the graphical analogy of the interaction effects beneath (Figures two, 3) show, predictions outside the range of 0 and i do not appear to be an issue here. Fourth, another argument against the LPM is that heteroscedasticity is most inevitably present. For this reason and to account for the sampling (see below), we estimate robust standard errors. Nonetheless, to scrutinize the robustness of our analyses, we take additionally estimated standard logit models with cluster robust standard errors applying the same weights every bit for the LPMs (see Table A4 in the Appendix).

Clustering, Standard Errors, and Weighting

PISA normally recommends to use balanced repeated replications (BRR) to estimate a coefficient's variance to take into account its circuitous sampling (OECD, 2009a; Lopez-Agudo et al., 2017). The particular variant used is known as Fay's method (Rust and Rao, 1996; Wolter, 2007). BRR breaks upwards the sample into subsamples ("replicates") and the estimate of interest is first estimated for the total sample and and so for each of the subsamples (Teltemann and Schunck, 2016). The estimator'south variance is so estimated as the differences betwixt the gauge from the full sample and each of the subsamples. We refrain from using BRR in this paper, because applying BRR may lead to a serious underestimation of the standard errors of country-level variables. Due to the resampling procedure, there volition exist no differences between the estimates for a land level variable in the full sample and the subsamples, because all students from 1 country have the aforementioned values for their state level variables.

Since the data is hierarchically structured with 3 clusters, it is necessary to business relationship for the iii-way clustering to estimate correct standard errors. Thus, we gauge cluster robust standard errors that account for the clustering at the country, the country-year, and the school level (Correia, 2017). Cluster-robust standard errors have shown to provide similar results for the lower level estimates when compared to BRR (Lopez-Agudo et al., 2017). To account for the complex sampling of PISA and the national differences in sampling, all analyses accept been weighted by normalized pupil weights. In contrast to the terminal student weight, which is recommended for inside-land analyses, applying these weights ensures that each country contributes equally to the analysis regardless of its actual size or student population.

Results

Figure ane shows the unadjusted risks for low operation among the unlike groups across the 30 countries in our sample averaged across 2009 and 2015. Nosotros see that beginning generation immigrants take a college risk of performing below the baseline level of reading proficiency than non-immigrant students in well-nigh countries of our sample.

Proportion of students below baseline level in reading, PISA, 2009 and 2015.

Beginning generation immigrant students also have a higher gamble of not reaching the baseline reading competence than second-generation immigrants in all countries except three (Republic of chile, Czech Republic, New Zeeland). Second generation students generally still have higher risks of depression performance compared to non-immigrants students with five exceptions (Australia, Canada, Israel, Hungary, Portugal), in which they bear witness similar or lower risks than their fellow non-immigrant students.

Table 2 gives the results of our multivariate analyses. Model ane includes just immigrant status and the country-level controls. It shows that starting time generation immigrants have a 16.1 percent points college probability of performing below the baseline level of proficiency than non-immigrants. Second generation immigrants take a eight.5 percentage points higher probability of low-performance than non-immigrants. After controlling for the individual-level characteristics (Model 2), the relatively higher hazard for immigrants is reduced: Second generation immigrants only have about two per centum points higher risk of performing below the baseline level than non-immigrants, starting time generation immigrants yet have about 9 percentage points higher chance. Model 3 includes the time-varying measure for the proportion of students attending schools that regularly employ standardized tests. While the estimated association is negative, statistical uncertainty is too high—the effect is not statistically significant. We also exercise not observe statistically significant associations between the use of regular standardized tests and students' migration background (Model 4).

Table 2

Iv level linear probability models predicting not reaching reading level 2.

	ane	two	3	4	5	6	7	8
	b/se	b/se	b/se	b/se	b/se	b/se	b/se	b/se
Pupil level
Native	ref.	ref.	ref.	ref.	ref.	ref.	ref.	ref.
First generation	0.161***	0.089***	0.089***	0.153***	0.089***	0.154***	0.089***	0.212***
	(0.021)	(0.017)	(0.017)	(0.037)	(0.017)	(0.019)	(0.017)	(0.039)
Second generation	0.085***	0.020	0.020	0.060	0.020	0.071***	0.020	0.107**
	(0.015)	(0.014)	(0.014)	(0.035)	(0.014)	(0.017)	(0.014)	(0.034)
Gender [1 = female]		−0.092***	−0.092***	−0.092***	−0.092***	−0.092***	−0.092***	−0.092***
		(0.005)	(0.005)	(0.005)	(0.005)	(0.005)	(0.005)	(0.005)
Language of examination spoken at home		−0.094***	−0.094***	−0.093***	−0.093***	−0.095***	−0.094***	−0.094***
		(0.013)	(0.013)	(0.013)	(0.013)	(0.012)	(0.013)	(0.012)
Parental education
None		0.056**	0.055**	0.055**	0.058**	0.058**	0.055**	0.055**
		(0.021)	(0.021)	(0.021)	(0.022)	(0.022)	(0.021)	(0.021)
ISCED 1		ref.	ref.	ref.	ref.	ref.	ref.	ref.
ISCED 2		−0.038	−0.039	−0.040	−0.038	−0.039	−0.040	−0.040
		(0.023)	(0.023)	(0.023)	(0.024)	(0.023)	(0.023)	(0.023)
ISCED 3b,c		−0.111***	−0.112***	−0.113***	−0.110***	−0.111***	−0.113***	−0.112***
		(0.025)	(0.025)	(0.025)	(0.026)	(0.025)	(0.025)	(0.025)
ISCED 3a,iv		−0.152***	−0.153***	−0.154***	−0.151***	−0.152***	−0.153***	−0.153***
		(0.023)	(0.023)	(0.023)	(0.024)	(0.024)	(0.023)	(0.023)
ISCED 5b		−0.161***	−0.162***	−0.163***	−0.159***	−0.160***	−0.162***	−0.161***
		(0.023)	(0.023)	(0.022)	(0.023)	(0.023)	(0.023)	(0.023)
ISCED 5a,6		−0.176***	−0.177***	−0.177***	−0.175***	−0.175***	−0.177***	−0.176***
		(0.023)	(0.023)	(0.023)	(0.024)	(0.024)	(0.023)	(0.023)
Index of family wealth possessions		0.001	0.002	0.002	0.001	0.001	0.002	0.002
		(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
Index of cultural possessions		−0.040***	−0.040***	−0.040***	−0.040***	−0.040***	−0.040***	−0.040***
		(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
Index of domicile educational resource		−0.042***	−0.042***	−0.042***	−0.042***	−0.042***	−0.042***	−0.042***
		(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
State-yr level
Gross domestic product growth (almanac, %)	−0.000	−0.001	−0.001	−0.001	−0.002	−0.002	−0.001	−0.001
	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.002)	(0.002)
Education expenditure (% of GNI)	0.016	−0.008	−0.009	−0.009	−0.017	−0.016	−0.005	−0.003
	(0.015)	(0.012)	(0.012)	(0.012)	(0.012)	(0.012)	(0.019)	(0.019)
Migrant stock (% of population)	0.000	0.003	0.003	0.003	0.002	0.001	0.003	0.002
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Unemployment (%) amidst foreign born	0.001	0.000	0.000	0.000	0.001	0.001	0.000	0.000
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Proportion of educatee attention schools that			ref.	ref.	ref.	ref.	ref.	ref.
Regularly use mandatory stand. tests			−0.038	−0.033	−0.040	−0.041	−0.036	−0.036
			(0.036)	(0.035)	(0.033)	(0.034)	(0.035)	(0.035)
Prop. of schools X outset gen.				−0.085
				(0.056)
Prop. of schools Ten second gen.				−0.053
				(0.055)
Post achievement data publicly					−0.158*	−0.144*
					(0.067)	(0.069)
Achievement data publicly X first gen.						−0.160***
						(0.036)
Accomplishment data publicly X second gen.						−0.124***
						(0.029)
Provide adm. dominance with accomplishment information							0.029	0.051
							(0.101)	(0.099)
Achievement data adm. authorization Ten first gen.								−0.179**
								(0.055)
Achievement data adm. authority 10 second gen.								−0.125*
								(0.049)
Country and twelvemonth stock-still effects	Yes	Aye	Yeah	Yes	Aye	Yes	Yes	Yes
Abiding	0.065	0.453***	0.488***	0.481***	0.600***	0.593***	0.443*	0.437*
	(0.067)	(0.070)	(0.081)	(0.081)	(0.087)	(0.088)	(0.174)	(0.172)
N countries	30	xxx	30	xxx	30	30	30	30
Due north country–years	54	54	54	54	54	54	54	54
N schools	12,255	12,255	12,255	12,255	12,255	12,255	12,255	12,255
Due north students	422,172	422,172	422,172	422,172	422,172	422,172	422,172	422,172

In Models five and seven, accountability in terms of the provision of aggregated achievement information of schools to the general public (Model 5) or to administrative authorities (Model seven) is tested. Making accomplishment data bachelor to the public is associated with a reduced probability of depression reading performance among all students (b = −0.158, s.eastward. = 0.067, Model 5), while providing achievement information to administrative authorities is not associated with low reading performance (b = 0.029, s.due east. = 0.101, Model 7). These findings thus simply partly ostend the start hypothesis derived from the principal-agent framework.

Models vi and eight test the 2nd hypothesis, which states that the communication of examination results is expected to be associated with a reduced risk of depression performance particularly among immigrant students. To facilitate interpretation, the Figures ii, ​ three graphically display the interaction effects. The left y-axis shows the predicted probability of low performance based on the respective regression model. The scale of the left y-axis for each figure runs from 0.0 to 0.five; the figures thus encompass a range of fifty% points. The x-axis displays the proportion of students attending schools inside a land which provide achievement data to the general public (or an administrative authority). The background of each figure additionally shows a histogram of the empirical distribution of the country-year level variable, that is the proportion of students that attend schools which provide information about achievement information to the corresponding recipient; this relates to the correct y-axis. We limited the predicted values to an empirically reasonable range on the x-axis, i.due east., for which we accept observations in the information.

Probability of low performance co-ordinate to accountability (data posted publicly).

Probability of low performance according to accountability (data provided to authoritative authorities).

Figures 2, ​ three evidence a similar pattern: The more prevalent accountability is in a land, the lower is the risk of low performance amid immigrant students. Figure 2 shows a negative association between the public provision of aggregated achievement data and the run a risk of low reading performance for all students. The association is strongest for beginning generation immigrant students, reducing the take chances of low performance by about 20 percentage points beyond the range of 10. Figure 3 displays the estimated associations between the provision of aggregated accomplishment data to administrative regime and the risk of low reading functioning. At that place is a comparatively small effect for kickoff generation immigrant students, about nine percentage points across the range of x. While the association is also negative for second generation immigrant students, statistical incertitude is high, as indicated past the large confidence intervals. The association for non-immigrant students appears slightly positive, merely is far from statistical significance. Thus, the results are more often than not uniform with our second hypothesis.

Robustness Check

To see if the results of the analyses are sensitive to the modeling approach, we accept estimated 2 sets of additional models. First, we have re-estimated all models as logit models with country and wave fixed effects and cluster robust standard errors, using the same weights as in the LPMs (see Tabular array A4 in the Appendix). The results of the logit models support the conclusions fatigued from the LPMs, with regard to the management of the relevant coefficients and their statistical uncertainty. The logit models, besides, approximate statistically significant, negative interaction effects, indicating that the provision of aggregated achievement data to the general public or to authoritative authorities is associated with a reduced probability of low reading achievement amid immigrant, in particular first generation, students. Equally in the LPMs, standardized testing alone is not statistically significantly associated with the risk of depression reading performance—neither for immigrant nor for non-immigrant students. 2d, we have re-estimated the models with the cross-level interaction as random effect models (with fourth dimension fixed effect) and included random slopes for the interaction term. This may be necessary as leaving out a random slope for a cross-level interaction may crusade the standard errors to exist biased downwards (Heisig and Schaeffer, 2019). The results (see Table A5 in the Appendix) besides back up the conclusions drawn from the LPM. The provision of aggregated accomplishment information to the public or to administrative authorities is associated with lower probability of low reading achievement for immigrant students. Even so, statistical uncertainty for the latter association is too loftier, i.e., the interaction furnishings are not statistically significant.

Word and Outlook

In this paper, nosotros examined the effects of standardized testing and the publication of school achievement data on low reading functioning for immigrant and non-immigrant students in 30 OECD countries using a longitudinal design at the country level by combining OECD PISA data from 2009 and 2015. Nosotros conceptualized depression performance every bit the take chances of performing below the and then-called baseline level of reading proficiency in the PISA study (OECD, 2016, p. 164). With respect to immigrant students and their prospects for societal integration, operation above this baseline level is crucial, as information technology measures 1'due south power to fully participate in a guild (OECD, 2009b, p. 2). Nosotros aimed at providing a more direct exam for arguments drawn from the master-agent models (William and Michael, 1976; Ferris, 1992; Laffont and Martimort, 2002), which are often mentioned in enquiry on standardized testing and educational performance (Wößmann, 2005) merely rarely directly tested.

Drawing on arguments from said principal-agent models, nosotros hypothesized that standardized testing itself should not be sufficient to prevent depression performance of students. Nosotros argued that an effect would merely sally if the principal, i.e., the administrative regime or parents, had access to results of such testing. This would alleviate the information asymmetry betwixt principal and agent, creating incentives for the agent (i.e., the school or the student) to prevent low performance. We furthermore expected immigrant students to profit more from this course of accountability than not-immigrant students, every bit they are often in need of special support.

The results of our analyses of PISA 2009 and 2015 reading data show that commencement, the use of standardized achievement tests solitary was not associated with the risk of low operation. Second, making the results of standardized tests available to the public was associated with a decreased risk of depression reading operation amongst all students, and third, particularly amid kickoff generation immigrant students. While the analyses as well tended to ostend this relationship if the testing results were made bachelor to an administrative authority, the estimated associations were smaller and not as robust. In a nutshell, the higher the share of schools that provide achievement data to the public, the lower is the chance for students, in particular for offset generation immigrant students, to perform below reading level 2. These results were robust across the three modeling approaches we used: linear probability multilevel models with country and yr furnishings and adapted standard errors for multiple clustering (Wooldridge, 2010; Correia, 2017), linear probability models with year fixed effects and random slopes for the cross-level interactions (Heisig and Schaeffer, 2019) and cluster robust standard errors, as well as logit models with land and year fixed effects and cluster robust standard errors.

Overall, the results supported the hypotheses drawn from the principal-agent-model, as they showed that the mere beingness of regular assessments is non sufficient to mitigate the information asymmetry between chief and agent if information from these assessments is not attainable. Assessments thus have to exist combined with adequate measures of accountability in order to incentivize the actors to align their efforts with the chief'south goals. The effects of assessments and accountability go especially credible in the context of low performance and in detail for a specific group: immigrant students. We argued that assessments, which are oft geared toward ensuring minimal levels of education, increase the incentives to support students at risk. As sufficient didactics is key for immigrant integration, education policies which lower the take a chance of depression performance gain in importance.

Limitations

Our written report has several limitations that should be considered. First, the forcefulness of international comparisons equally nosotros conducted it, is the variation in institutional characteristics. However, although all countries belong to the OECD, they are still heterogenous not the least with respect to their clearing history, which may be confounded with both educational institutions and (immigrant) student operation. We tried to arroyo this trouble with a longitudinal arroyo at the country level, finer controlling for all time-constant differences between countries, by focusing only on changes in the institutional arrangements within countries over time. Withal, nosotros only have two measurements over time. What is more, although nosotros have tried to include the about relevant time-varying confounders at the country-yr level, the estimated results are even so prone to bias due to unobserved heterogeneity. Larger fourth dimension-spans and additional meaningful controls at the land-year level would strengthen the analytical pattern. Second, it is unfortunate that PISA does non allow for a systematic and comparable differentiation of immigrant origin. We have attempted to alleviate this problem partially by controlling for different aspects of parental socio-economical status and linguistic communication use at home. Yet, we have to wait that the overall effect that we observed volition vary across dissimilar countries of origin. Still, the association is clearly present, even if the effect may exist heterogenous beyond immigrant groups. Third, we have called a four-level linear probability model to clarify the information for the reasons outlined in the Data and Methods department, since the potentially amend suited model (iv level logit) could not be used. Nevertheless, comparisons of the LPMs' results with other modeling approaches (single level logit models and random intercept random slope models) showed very similar results. This increases our confidence that the results are not artifacts of the modeling approach. 4th, the main proportion of variance in educational performance, including the take chances of low operation, lies at the individual level. If we audit empty random effect models, the intra-class correlations for the state and the country-yr level are estimated to being but effectually 0.03. This has to be taken into consideration, when evaluating the results. The low intra-class correlation could be seen equally an argument against investigating characteristics at the country(-twelvemonth) level. Clearly, individual factors are responsible for the larger share of variation in educational performance. Yet, nosotros call back that it is however relevant to analyze the function of institutional characteristics. From a policy perspective, institutional regulations are easier to adjust than students' characteristics. In a short term perspective, the latter has to be seen given. Profound cognition about the furnishings—admitting minor—of institutional characteristics of education organisation is crucial if one is interested in shaping institutions which facilitate sustainable development and organisation integration of contemporary societies. Fifth, although we tried to put the propositions of the principlal-agent framework to a direct test, nosotros nonetheless face a black-box. With the data at hand, we do not know for certain if the mechanisms that create the association between (immigrant) student achievement and the public provision of cess data correspond to those outlined in the principal agent framework. Further enquiry could attempt to out even more than specific hypotheses to the examination. Our analyses neglect to falsify predictions from the model, simply should not be seen as a proof that the model is correct.

In summary, our results bear witness that the mere implementation of standardized assessments has no effects on low reading performance, neither for immigrant nor for non-immigrant students. In line with the predications from a primary-agent framework, nosotros do find a general association betwixt provision of assessment data to the public and the risk for low reading functioning. Start generation immigrant students in particular take a reduced probability for low reading performance in countries that make cess data available publicly.

Author Contributions

JT and RS take jointly conceptualized and drafted the manuscript, canonical information technology for publication, conceptualized the research question, and the theoretical arroyo. JT has conducted the literature review. RS conducted the empirical analyses. All authors contributed to the article and approved the submitted version.

Conflict of Interest

The authors declare that the research was conducted in the absenteeism of any commercial or financial relationships that could exist construed as a potential disharmonize of interest.

Acknowledgments

We thank Katja Pomianowicz for helpful comments on an earlier version of this manuscript.

Footnotes

^aneOne might argue that the signaling machinery that is often referred to in the literature is very specific (Bishop, 1995). However, we practice not know of whatever study using large scale assessment data, similar PISA, which explicitly tests the mechanism, that is investigating if students actually attach more value or importance to their pedagogy in the presence of standardized exit exams.

²https://www.oecd.org/pisa/data/

³Every bit the sampling design of PISA targets the student population, non the schools in a land, ciphering of state level variables by aggregation has to be done with the (weighted) student level data. Since the sampling frames in PISA aim to provide representative information on all eligible students within a country, the resulting variables measure the proportion of students in a country attending schools with a corresponding characteristic (e.g., schools that make cess data publicly available).

References

• Bergbauer A., Hanushek E. A., Wößmann L. (2018). Testing. NBER Working Paper 24836. 10.3386/w24836 [CrossRef] [Google Scholar]
• Berry W. D., Golder 1000., Milton D. (2012). Improving tests of theories positing interaction. J. Polit. 74, 653–671. 10.1017/S0022381612000199 [CrossRef] [Google Scholar]
• Bishop J. H. (1995). The bear on of curriculum-based external examinations on school priorities and pupil learning. Int. J. Educ. Res. 28, 653–752. ten.1016/0883-0355(96)00001-8 [CrossRef] [Google Scholar]
• Bishop J. H. (1997). The result of national standards and curriculum-based exams on accomplishment. Am. Econ. Rev. 87, 260–264. [Google Scholar]
• Bodovski K., Byun Due south., Chykina Five., Chung H. J. (2017). Searching for the gilded model of teaching: cross-national analysis of math achievement. Compare 47, 722–741. 10.1080/03057925.2016.1274881 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Bol T., van de Werfhorst H. G. (2013). The Measurement of Tracking, Vocational Orientation, and Standardization of Educational Systems: A Comparative Approach. Gini Hash out. Pap. 81. [Google Scholar]
• Bol T., Witschge J., van de Werfhorst H. G., Dronkers J. (2014). Curricular tracking and cardinal examinations: counterbalancing the impact of social background on student achievement in 36 countries. Soc. Forces 92, 1545–1572. x.1093/sf/sou003 [CrossRef] [Google Scholar]
• Booher-Jennings J. (2005). Below the bubble: "Educational triage" and the texas accountability system. Am. Educ. Res. J. 42, 231–268. 10.3102/00028312042002231 [CrossRef] [Google Scholar]
• Brambor T., Clark West. R., Golder M. (2006). Understanding interaction models: improving empirical analyses. Polit. Anal. 14, 63–82. 10.1093/pan/mpi014 [CrossRef] [Google Scholar]
• Carnoy M., Loeb Due south. (2002). Does external accountability affect educatee outcomes? A cantankerous-land analysis. Educ. Eval. Policy Anal. 24, 305–331. ten.3102/01623737024004305 [CrossRef] [Google Scholar]
• Chmielewski A. K., Reardon S. F. (2016). Patterns of cross-national variation in the association between income and academic achievement. AERA Open 2:233285841664959. 10.1177/2332858416649593 [CrossRef] [Google Scholar]
• Cobb-Clark D. A., Sinning Thou., Stillman Due south. (2012). Migrant youths' educational achievement: the role of institutions. Ann. Am. Acad. Politico. Soc. Sci. 643, 18–45. 10.1177/0002716212440786 [CrossRef] [Google Scholar]
• Correia Southward. (2017). Linear models with high-dimensional stock-still effects: an efficient and feasible estimator, in Working Paper, 1–18. Available online at: http://scorreia.com/research/hdfe.pdf
• Ferris J. M. (1992). Schoolhouse-based decision making: a main-agent perspective. Educ. Eval. Policy Anal. 14, 333–346. 10.3102/01623737014004333 [CrossRef] [Google Scholar]
• Fontenay S. (2018). sdmxuse: command to import data from statistical agencies using the SDMX standard. Stata J. Promot. Commun. Stat. Stata xviii, 863–870. 10.1177/1536867X1801800407 [CrossRef] [Google Scholar]
• Fuchs T., Wößmann Fifty. (2007). What accounts for international differences in educatee performance? A re-examination using PISA data. Empir. Econ. 32, 433–464. 10.1007/s00181-006-0087-0 [CrossRef] [Google Scholar]
• Heisig J. P., Schaeffer Thousand. (2019). Why you should always include a random gradient for the lower-level variable involved in a cross-level interaction. Eur. Sociol. Rev. 35, 258–279. 10.1093/esr/jcy053 [CrossRef] [Google Scholar]
• Hellevik O. (2009). Linear versus logistic regression when the dependent variable is a dichotomy. Qual. Quant. 43, 59–74. ten.1007/s11135-007-9077-iii [CrossRef] [Google Scholar]
• Horn D. (2009). Age of pick counts: a cross-country analysis of educational institutions. Educ. Res. Eval. xv, 343–366. ten.1080/13803610903087011 [CrossRef] [Google Scholar]
• Jensen Chiliad. C., Meckling Due west. H. (1976). Theory of the firm: managerial beliefs, agency costs and ownership structure. J. Financ. Econ. three, 305–360. ten.1016/0304-405X(76)90026-X [CrossRef] [Google Scholar]
• Jürges H., Schneider K., Büchel F. (2005). The effect of primal exit examinations on student achievement: quasi-experimental evidence from TIMSS Germany. Eur. Econ. Assoc. 3, 1134–1155. 10.1162/1542476054729400 [CrossRef] [Google Scholar]
• Klieme E., Avenarius H., Blum W., Döbrich P., Gruber H., Prenzel Grand., et al. . (2003). ZurEntwicklung nationaler Bildungsstandards, sogen. Klieme-Expertise. Berlin: BMBF. Available online at: www.dipf.de/publikationen/volltexte/zur_entwicklung_nationaler_bildungsstandards.pdf (accessed November 05, 2020). [Google Scholar]
• Laffont J.-J., Martimort D. (2002). The Theory of Incentives: The Primary-Agent Model. Princeton, NJ: Princeton University Press. 10.1515/9781400829453 [CrossRef] [Google Scholar]
• Levačić R. (2009). Teacher incentives and performance: an application of principal–agent theory. Oxford Dev. Stud. 37, 33–46. 10.1080/13600810802660844 [CrossRef] [Google Scholar]
• Long J. Southward. (1997). Regression Models for Categorical and Limited Dependent Variables. Chiliad Oaks, CA: Sage Publication. [Google Scholar]
• Lopez-Agudo Fifty. A., Jerrim J., Marcenaro Gutierrez O., Shure Due north. (2017). To weight or not to weight? the case of PISA information, in BT–Investigaciones de Economía de la Ed. eds Gallego J. C. M., Cárceles Chiliad. C. P., Torrejón L. Due north. (Asociación de Economía de la Educación), 285–302. Available online at: https://econpapers.repec.org/RePEc:aec:ieed12:12-xiii (accessed November 05, 2020).
• Macdonald Thousand. (2019). PV: stata module to perform estimation with plausible values. Stat. Softw. Components S456951. [Google Scholar]
• Meyer H.-D., Benavot A. (2013). PISA, Power, and Policy: The Emergence of Global Educational Governance: Oxford Studies in Comparative Teaching. 2d Edn. Oxford: OECD PISA Information Analysis Manual: SPSS. 10.15730/books.85 [CrossRef] [Google Scholar]
• Mittleman J., Jennings J. 50. (2018). Accountability, achievement, and inequality in american public schools: a review of the literature, in Handbook of the Sociology of Education in the 21st Century (Cham: Springer; ), 475–492. 10.1007/978-3-319-76694-2_21 [CrossRef] [Google Scholar]
• Motiejunaite A., Noorani S., Monseur C. (2014). Patterns in national policies for back up of low achievers in reading beyond Europe. Br. Educ. Res. J. forty, 970–985. x.1002/berj.3125 [CrossRef] [Google Scholar]
• OECD (2009a). PISA Data Assay Manual: SPSS, 2nd Edn. Paris: OECD Publishing. 10.1787/9789264056275-en [CrossRef] [Google Scholar]
• OECD (2009b). Take the Test: Sample Questions from OECD'southward PISA Assessments. Paris: OECD Publishing. [Google Scholar]
• OECD (2010). Pathways to Success: How Knowledge and Skills at Age 15 Shape Futurity lives in Canada. Paris: OECD. 10.1787/9789264081925-en [CrossRef] [Google Scholar]
• OECD (2013). PISA 2012 Results: What Makes Schools Successful? Resource, Policies and Practices (volume IV). Paris: OECD Publishing. [Google Scholar]
• OECD (2016). PISA 2015 Results (Volume I): Excellence and Equity in Educational activity. Paris: OECD Publishing. [Google Scholar]
• OECD (2017). PISA 2015 Technical Report. Paris: OECD Publishing. [Google Scholar]
• Rust K., Rao J. (1996). Variance interpretation for complex surveys using replication techniques. Stat. Methods Med. Res. v, 283–310. 10.1177/096228029600500305 [PubMed] [CrossRef] [Google Scholar]
• Scheerens J. (2007). The case of evaluation and accountability provisions in education as an surface area for the evolution of policy malleable arrangement level indicators, in Bildungs- und Sozialberichterstattung, eds Krüger H.-H., Rauschenbach T., Sander U. (Wiesbaden: VS Verlag für Sozialwissenschaften; ), 207–224. ten.1007/978-3-531-90615-7_16 [CrossRef] [Google Scholar]
• Schneeweis N. (2011). Educational institutions and the integration of migrants. J. Popul. Econ. 24, 1281–1308. ten.1007/s00148-009-0271-6 [CrossRef] [Google Scholar]
• Schütz G., West 1000., Wößmann Fifty. (2007). School Accountability, Autonomy, Choice, and the Disinterestedness of Educatee Achievement: International Evidence From PISA 2003. OECD Education Working Papers, No. fourteen. Paris: OECD Publishing. 10.1787/246374511832 [CrossRef] [Google Scholar]
• Shipley L., Gluzynski T. (2011). Life-path outcomes at age 25 associated with reading ability at age 15, in Pedagogy Matters: Insights on Didactics, Learning and Grooming in Canada, Vol. 8 (Statistics Canada periodicals). [Google Scholar]
• Teltemann J. (2015). Ungleichheit als Organization? Der Schulerfolg von Migranten im internationalen Vergleich. Frankfurt/Main: Campus. [Google Scholar]
• Teltemann J., Schunck R. (2016). Education systems, schoolhouse segregation, and second-generation immigrants' educational success: evidence from a country-fixed effects approach using three waves of PISA. Int. J. Comp. Sociol. 57, 401–424. 10.1177/0020715216687348 [CrossRef] [Google Scholar]
• William H., Michael J. (1976). Theory of the firm: managerial beliefs, agency costs and ownership structure. J. Financ. Econ. 3, 305–360. [Google Scholar]
• Wolter 1000. (2007). Introduction to Variance Estimation, 2d Edn. New York, NY: Springer [Google Scholar]
• Wooldridge J. M. (2010). Econometric Analysis of Cross Section and Console Data. Cambridge, MA: MIT Printing [Google Scholar]
• Wößmann L. (2003). Schooling resources, educational institutions and student performance: the international prove. Oxf. Bull. Econ. Stat. 65, 117–170. 10.1111/1468-0084.00045 [CrossRef] [Google Scholar]
• Wößmann L. (2005). The effect heterogeneity of cardinal exams: bear witness from TIMSS, TIMSS-repeat and PISA. Educ. Econ. xiii, 143–169. 10.1080/09645290500031165 [CrossRef] [Google Scholar]
• Wößmann L., Lüdemann E., Schütz One thousand., Westward M. (2007). Schoolhouse Accountability, Autonomy, Choice, and the Level of Student Achievement: International Evidence from PISA 2003. Paris: OECD Publishing. [Google Scholar]
• Wößmann Fifty., Lüdemann E., Schütz G., W M. R. (2009). School Accountability, Autonomy, and Pick around the World. Cheltenham and Northhampton and MA: Edward Elgar. [Google Scholar]

---

Manufactures from Frontiers in Folklore are provided hither courtesy of Frontiers Media SA

---

gibbonswitiou.blogspot.com

Source: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8022478/

Share this post