Search

The main objective of scientific research is to solve problems. Empirical studies are affected by random and systematic errors. Random errors decrease the precision of study results, but do not bias these results. In contrast, systematic errors bias study results. Errors and methods to prevent and correct errors are introduced, and are put into the context of the different parts of empirical studies (i.e., research questions, literature review, sampling, operationalizations, design, implementation, data analysis, and reporting).

Pretest-posttest change can be studied at the population level and at the individual level. Within-group change is the change of a population parameter, for example, the mean from pretest to posttest. Between-groups change is the difference of within-group change between (e.g., E- and C-) groups. In contrast, within-person change is the change of a single person. In general, change at the population level does not apply to single persons. Within-group change is assessed by testing the null hypothesis of equal pretest and posttest population means. Between-groups change can be tested with different methods. The test based on posttest-pretest difference scores is described. Single-person change is derived from the participant’s observed test scores or his (her) item responses. Change methods based on item responses depend on the scale of the items. Methods for continuous and dichotomous item responses are described. The participant’s observed pretest and posttest scores are used to test the null hypothesis of no true score change. The same null hypothesis can be tested with the participant’s item responses if these responses are at a continuous scale. The method uses the participant’s continuous item responses instead of his (her) test scores. An individual change measure is defined for dichotomous item responses. This measure is applied to test the null hypothesis of no change of a single person. It is recommended to study pretest-posttest change at both the population and the individual level because these levels yield different information.

A factor is an independent variable. A factorial design completely crosses two or more factors. It is an efficient design that simultaneously studies main effects of factors and their interactions. A distinction is made between factors that can be manipulated by researchers and factors that cannot be manipulated. Usually, the effects of manipulable factors are causally interpreted. Nonmanipulable factors are included for two different reasons, but they should not be dichotomized. First, to increase the precision of parameter estimates and the power of statistical tests. The nonmanipulable factor is included as a blocking variable in a randomized block design (if its values are known before participants are assigned to conditions) or as a covariate in the analysis of the data. Second, to study their relations with manipulable factors, but these relations should not be causally interpreted. The statistical analysis of factorial design data has to be tuned to the type of dependent variable (DV). Usually, ANOVA or ANCOVA are applied to (approximately) continuous DVs, but these methods make strong assumptions. Akritas et al.’s (J Am Stat Assoc 92:3375–3384, 1997) nonparametric method is more appropriate to analyze (approximately) continuous and ranked DVs. The preferred methods for dichotomous, nominal-categorical, and ordinal-categorical DVs are the logit, baseline-category, and cumulative logit models, respectively. Often, researchers apply omnibus statistical tests to factorial design data. These tests mainly fit into exploratory research. Confirmatory studies prespecify specific hypotheses. The proper methods to test these hypotheses are planned comparisons of conditions.

Null hypothesis testing applies to confirmatory research, where substantive hypotheses are tested. The preferred approach is to construct a confidence interval (CI) because a CI simultaneously assesses the precision of a parameter estimate and tests a null hypothesis on the parameter. Two- and one-sided CIs and two- and one-tailed tests are considered. The CI approach is demonstrated for conventional tests of the null hypothesis of equal means of paired (Student’s t test) and independent (Student’s t and Welch tests) variables. Bootstrap methods make weaker assumptions than the conventional tests. The bootstrap t method for the means of paired and independent variables and the modified percentile bootstrap method for the product moment correlation are described. Null hypothesis testing is often incorrectly understood and applied. Several methods to correct these flaws are discussed. First, the overlap of the CIs of two means does not imply that the difference of the two means is not significant. Second, a two-step procedure, where the choice of a test is based on results of tests of the assumptions of the test, inflates the Type I error. Third, standardized effect sizes can be computed in different ways, which hampers the comparability of effect sizes in meta-analysis. Fourth, an observed power analysis, where the effect size is estimated from sample data, cannot explain nonsignificant results. Fifth, testing multiple null hypotheses increases the probability of rejecting at least one true null hypothesis, which is prevented by applying multiple null hypothesis testing methods (e.g., Hochberg’s method). Sixth, data exploration may yield interesting substantive hypotheses, but these have to be confirmed with new data of a cross-validation or replication study. Seventh, adding participants to the sample till the null hypothesis is rejected inflates the Type I error, which is prevented by using sequential testing methods (e.g., the group sequential testing procedure). Finally, if researchers do not want to reject a null hypothesis, they have to apply equivalence testing.

A bivariate relation is a relation between two (X and Y) variables. A large number of coefficients for bivariate relations was developed. A classification of coefficients is presented to facilitate the choice of an appropriate coefficient. The classification is based on two distinctions. First, three types of relations are distinguished: (1) symmetrical (the relation between X and Y is the same as between Y and X), (2) equality (symmetrical and equality of X- and Y-values), and (3) asymmetrical (one variable is the independent variable or predictor and the other variable is the dependent variable or criterion). Second, five types of variables are distinguished: (1) dichotomous (two unordered or ordered categories), (2) nominal-categorical (more than two unordered categories), (3) ordinal-categorical (more than two ordered categories), (4) ranked (rank numbers), and (5) continuous (values from a continuum). Crossing of these two distinctions yields 3 × 5 = 15 different combinations. An example of a coefficient is given for 13 of these combinations (coefficients for the two other combinations are not known to the author). The examples are restricted to coefficients for relations between two variables of similar type, for example, Cohen’s kappa for an equality relation between two nominal-categorical variables.

Standardized effect sizes are frequently used, for example, in sample size planning and meta-analysis. However, they are often hard to interpret, and can be estimated in different ways. A measure that is better to interpret than the standardized difference of two means is the probability of superiority. The probability of superiority of paired (e.g., pretest and posttest) scores is the probability that of a randomly selected pair of scores the second (e.g., posttest) score is larger than the first (e.g., pretest) score. The probability of superiority of independent (e.g., E- and C-group) scores is the probability that a randomly selected participant from one (e.g., The E-) group has a larger score than a randomly selected participant from the other (e.g., C-) group. The interpretability of observed test scores is facilitated by applying linear transformations to the scores. Two transformations are described. The Average Item Score (AIS) is the mean of the item scores of the test. The Proportion of Maximum Possible (POMP) score is the proportion that the observed test score takes of the distance between the minimum and maximum possible scores of the test.

A psychological or educational test is an instrument for the measurement of a person’s maximum performance or typical response under standardized conditions, where the performance or response is assumed to reflect one or more latent variables. A test consists of a set of items. Conventional test scoring assigns a priori scores to test takers’ item responses, and a test taker’s observed test score is the sum of his (her) item scores. Test scores are affected by random and systematic errors. Random errors decrease the measurement precision of tests, and systematic errors bias the measurements. A within-person and a between-persons aspect of measurement precision are distinguished. The within-person aspect is the variance of a given test taker’s observed score across hypothetical replications, which assesses the precision of the measurement of the test taker’s true score. The between-persons aspect is the reliability, which is the squared product moment correlation between observed and true test scores in a population of test takers. Measurement precision is increased by applying guidelines for test construction and administration. Classical and modern psychometric methods assess the quality of tests and items. Classical item analysis indices are the item p-value and item-rest correlation, and modern indices are the item difficulty and discrimination parameters of item response models.

Systematic errors bias measurements. Obviously, cheating is a biasing factor. Answer copying is a type of cheating, which can be detected by answer-copying indices. Satisficing is responding to items with less than optimal efforts. It is detectable by building instructional manipulation checks or bogus items into the test. A response style is a typical way of taking a test. Plodding is a style where test takers start slowly and take much time to answer items. It is counteracted by informing test takers on the amount of time and number of items that remain. Fumbling is a style where test takers start rather panicky. It cannot be prevented, but testing should be less panic arousing as possible. The extremity and midpoint styles are the tendencies to choose extreme and midpoint answer categories, respectively. Acquiescence and dissentience are tendencies of yea- and nay-saying, respectively, and are detectable if the test is balanced (i.e., has about equal numbers of indicative and contra-indicative items). Satisficing and fumbling yield aberrant response patterns and acquiescence and dissentience aberrant response patterns at balanced tests that are detectable by person-fit indices. Item nonresponse is the tendency to skip items. It is reduced by computerized test administration and re-approaching test takers who omitted items. These styles and tendencies are sometimes person characteristics that are of interest of their own. Deleting test takers who show styles or tendencies yields missing data that have to be handled in the analysis of the data (see Chap. 16).

A general linear model was described for tests constructed in a facet design. Guilford's structure of intellect model and an alternative model were specified as special cases. From the Guilford research seven data sets were selected such that both models could be compared. The comparison was done with oblique and orthogonal factors. Using covariance structure analysis the fit of the models was assessed, and the parameters estimated. Models with orthogonal factors did not fit the data. The fit of the oblique Guilford model was better than the alternative oblique model. For three of the data sets the oblique Guilford model yielded an acceptable fit to the data with parameter estimates that could be interpreted by the structure of intellect model.

A general linear latent trait model for continuous item responses is described. The special unidimensional case for continuous item responses is Joreskog's (1971) model of congeneric item responses. In the context of the unidimensional case model for continuous item responses the concepts of item and test information functions, specific objectivity, item bias, and reliability are discussed; also the application of the model to test construction is shown. Finally, the correspondence with latent trait theory for dichotomous item responses is discussed.

In a randomized experiment (n = 515), a computerized and a computerized adaptive test (CAT) are compared. The item pool consists of 24 polytomous motivation items. Although items are carefully selected, calibration data show that Samejima's graded response model did not fit the data optimally. A simulation study is done to assess possible consequences of model misfit. CAT efficiency was studied by a systematic comparison of the CAT with two types of conventional fixed length short forms, which are created to be good CAT competitors. Results showed no essential administration mode effects. Efficiency analyses show that CAT outperformed the short forms in almost all aspects when results are aggregated along the latent trait scale. The real and the simulated data results are very similar, which indicate that the real data results are not affected by model misfit.

Objective: To validate the QUALIDEM, a quality of life measure for people with dementia within residential settings rated by professional caregivers. Method: In a sample of 202 residents of nursing homes Spearman rank correlations were calculated between the QUALIDEM subscales aand indices of convergent validity and discriminant validity, with dementia severity and need of care, with global QOL scores by the head nurse and family, and with self-report on COOP/WONCA Charts. Results: The one-method multi-trait matrix showed 90.5% of the correlations to be in support for convergent and discriminant validity. Low to moderate correlations were observed with dementia severity and need of care, confirming that QOL is not merely disease severity. Support for concurrent validity was found in correlations with QOL ratings by the head nurse. The QUALIDEM did not correlate with most of the family ratings or with the COOP/WONCA Charts. Conclusion: The results of this validation study together with the obtained content validity through the method of construction provide sufficient support for validity of the QUALIDEM to be used for care evaluation and research in residential settings.

Objective: To develop a Quality of Life questionnaire rated by professionals that can be used for people with dementia in different stages of the disease, living in residential settings. Method: Development was performed in two phases: item generation and pilot testing, and a field survey to evaluate the psychometric properties. For unidimensionality we used a non-parametric model from item response theory: the Mokken scaling model, and computed the corresponding scalability coefficients, using a theory driven strategy. Results: The pilot survey resulted in a list of 49 items. The field survey was performed in a sample of 238 people with dementia residing in ten nursing homes. The scalability of the subscales positive affect, negative affect, restless tense behavior, and social relations is strong (0.50 < H < 0.63); for care relationship, positive self image, feeling at home, and having something to do, scalability was moderate (0.40 < H < 0.49), and for social isolation it was weak (H = 0.34). The reliability coefficient Rho (under assumption of double monotonicity) varied from 0.60 for social isolation to 0.90 for positive affect (Cronbach's alpha varied from 0.59 to 0.89). Twenty-one of 40 items are suited for people with very severe dementia. Conclusion: The QUALIDEM is an easy to administer and sufficiently reliable rating scale that provides a QOL profile of persons with dementia in residential settings. The QUALIDEM can be used for evaluation as well as for research and practice innovation.

Some breast cancer survivors experience cognitive decline following chemotherapy. We prospectively examined changes in cognitive performance among high-risk breast cancer patients who had received high-dose chemotherapy with cyclophosphamide, thiotepa, and carboplatin (CTC group; n = 28) or standard-dose chemotherapy with 5-fluorouracil, epirubicin, and cyclophosphamide (FEC group; n = 39); stage-I breast cancer patients who had received no systemic chemotherapy (no-CT group; n = 57); and healthy control subjects (n = 60). All patients underwent neuropsychologic testing before and 6 months after treatment (12-month interval); control subjects underwent repeated testing over a 6-month interval. No differences in cognitive functioning between the four groups were observed at the first assessment. More of the CTC group than the control subjects experienced a deterioration in cognitive performance over time (25% versus 6.7%; odds ratio [OR] = 5.3, 95% confidence interval [CI] = 1.3 to 21.2, P = .02). No such difference was observed for the FEC or the no-CT groups (FEC versus control: OR = 2.2, 95% CI = 0.5 to 9.1, P = .27; no-CT versus Control: OR = 2.2, 95% CI = 0.6 to 8.0; P = .21). Some cytotoxic treatment for breast cancer affects cognition in a subset of women.

A total of 520 high school students were randomly assigned to a paper-and-pencil test (PPT), a computerized standard test (CST), or a computerized adaptive test (CAT) version of the Dutch School Attitude Questionnaire (SAQ), consisting of ordinal polytomous items. The CST administered items in the same order as the PPT. The CAT administered all items of three SAQ subscales in adaptive order using Samejima’s graded response model, so that six different stopping rule settings could be applied afterwards. School marks were used as external criteria. Results showed significant but small multivariate administration mode effects on conventional raw scores and small to medium effects on maximum likelihood latent trait estimates. When the precision of CAT latent trait estimates decreased, correlations with grade point average in general decreased. However, the magnitude of the decrease was not very large as compared to the PPT, the CST, and the CAT without the stopping rule.

This paper comments on an article by Michell (2000), who argues that psychometrics should be qualified as pathological science for two reasons: (a) psychometrics assumes psychological attributes to be quantitative without testing this hypothesis; and (b) the fact that this hypothesis is not tested is disguised. Michell further argues that the hypothesis should be tested using additive conjoint measurement theory. Although relevant to classical test theory, Michell’s arguments do not apply to psychometrics in general. In particular, they are largely irrelevant to item response theory models. We show that these models result from introducing probabilistic relations, which are needed to deal with measurement error, and not from a breakdown in critical inquiry, as Michell suggests. Moreover, at least one class of these models can be formulated in terms of additive conjoint measurement theory, which renders Michell’s call for the additive conjoint model in need of qualification. Finally, item response theory models are routinely tested against empirical data, and although the assumption that an attribute is quantitative cannot be tested directly, such tests do address the conjunction of this assumption and other model assumptions. We conclude that, although Michell’s arguments are important to psychological measurement, they are largely irrelevant to item response theory. In fact, we argue that they can be phrased in terms of this theory in a natural way.

The study tested the hypotheses that integrated support, in which patients and carers are both supported by one professional staff member, will be more effective in reducing the feelings of (over)burden of carers and in positively influencing some potential determinants of experienced burden than non-integrated support, such as psychogeriatric day-care. A quasi-experimental pre-test/post-test control group design was applied with matched groups and measurements at baseline and after seven months. Fifty-five of the eighty dyads (31% dropped out) of persons with mild-to-moderate dementia and their carers who presented for four community meeting centres and three day-care centres in nursing homes over an 18 month period, were included in the study. The carers in the Meeting Centres Support (MCS) Program participated in an integrated family support program together with the persons with dementia, while the carers in the control group only received respite through psychogeriatric day-care. Standardized questionnaires were applied to measure some indicators of burden experienced by the carers (feelings of stress, life dissatisfaction, psychological and psychosomatic complaints), as well as some potential determinants of experienced burden (feeling of competence, coping strategies, experienced support and loneliness). Furthermore, as an indicator of feelings of (over)burden of the carer, the time between start of participation in one of the programs and institutionalization in a nursing home was calculated. After seven months the carers in the MCS group (n = 36), compared to the carers in regular psychogeriatric day-care (n = 19), showed a moderate positive effect on the feeling of competence (effect size 0.45). No effect was found on feelings of stress, dissatisfaction or psychological and psychosomatic complaints. However, as compared to the control group in regular day-care, the persons with dementia in the MCS Program participated for a longer period of time before they were placed in a nursing home. The MCS Program proved more effective than psychogeriatric day-care in influencing the feeling of competence of the carers, and seems to lead to an increased delay of nursing home placement of the person with dementia, as compared to regular day-care. Because of the small sample and limited power of the study, and the possible influence of the selection of carers on the study outcome, further research into the value of meeting centres for carers of people with dementia is recommended.

ObjectiveThis research studied the relationship between associated congenital malformations and the mental and psychomotor development of children with clefts.DesignThe study was cross-sectional.SettingThe study was conducted in a university hospital for children.ParticipantsThe sample consisted of 148 children with cleft lip, cleft palate, or both.Main Outcome MeasuresThe children were assessed by a clinical geneticist at the age of 18 months. The children's level of development was determined by means of the Dutch version of the Bayley Scales of Infant Development.ResultsOne-third of the total sample had associated malformations. Children with an isolated cleft lip showed the least. Children with an isolated cleft palate showed the highest percentage of minor malformations that are minor yet possibly worrisome. The total group achieved a mean developmental index (DI) on the mental scale of 98.9 with SD of 20.9. The motor scale showed a mean DI of 104.9 and SD of 24.7. Analysis of variance (ANOVA) showed that on the mental scale, the three main effects (diagnosis, evaluation, and sex) were significant at the 5% level. On the motor scale, only the main effect “evaluation” was significant.ConclusionsThis study demonstrated that children with associated congenital malformations might be disadvantaged with respect to their development. These malformations occurred most frequently with the cleft lip and palate and cleft palate only subgroups. More research, especially concerning the cleft palate only subgroup is needed because they are most at risk.

This article examines the theoretical status of latent variables as used in modern test theory models. First, it is argued that a consistent interpretation of such models requires a realist ontology for latent variables. Second, the relation between latent variables and their indicators is discussed. It is maintained that this relation can be interpreted as a causal one but that in measurement models for interindividual differences the relation does not apply to the level of the individual person. To substantiate intraindividual causal conclusions, one must explicitly represent individual level processes in the measurement model. Several research strategies that may be useful in this respect are discussed, and a typology of constructs is proposed on the basis of this analysis. The need to link individual processes to latent variable models for interindividual differences is emphasized.

Measurement bias refers to systematic differences across subpopulations in the relation between observed test scores and the latent variant underlying the test scores. Comparisons of subpopulations with the same score on the latent variable can be expected to have the same observed test score. Measurement invariance is therefore one of the key issues in psychological testing. It has been established that strict factorial invariance (SFI) with respect to a selection variable V almost certainly implies weak measurement invariance with respect to V: given SFI, means and variances of observed scores do not depend on V. It is shown that this result can be extended. SFI in groups derived by selection on V has implications not only for V but also for potentially biasing variables W, if W and the selection variable V and/or if W and the factor underlying the observed test scores are statistically dependent. Given SFI with respect to V and prior knowledge concerning these dependencies, it is not necessary to measure and model variables W in order to exclude them as potentially biasing variables if the investigation focuses on groups selected on V.

Investigating sources of within- and between-group differences and measurement invariance (MI) across groups is fundamental to any meaningful group comparison based on observed test scores. It is shown that by placing certain restrictions on the multigroup confirmatory factor model, it is possible to investigate the hypothesis that within- and between-group differences are due to the same factors. Moreover, the modeling approach clarifies that absence of measurement bias implies common sources of within- and between-group variation. It is shown how the influence of background variables can be incorporated in the model. The advantages of the modeling approach as compared with other commonly used methods for group comparisons is discussed and illustrated by means of an analysis of empirical data. © 2003 Elsevier Science Inc. All rights reserved.

In this article, a distinction is made between absolute and relative measurement. Absolute measurement refers to the measurement of traits on a group-invariant scale, and relative measurement refers to the within-group measurement of traits, where the scale of measurement is expressed in terms of the within-group position on a trait. Relative measurement occurs, for example, if an item induces a within-group comparison in respondents. These distinctions are discussed within the framework of measurement invariance, differentiating between absolute and relative forms of measurement invariance and bias. It is shown that items for relative measurement will produce bias as classically defined if the mean and/or variance of the trait distribution differ between groups. This form of bias, however, does not result from multidimensionality but from the fact that measurement is on a relative scale. A logistic regression procedure for the detection of relative measurement invariance and bias is proposed, as well as a model that allows for the incorporation of items for relative measurement in test analysis. Implications of the distinction between absolute and relative measurement are discussed and prove to be especially relevant for the domain of personality research.

This paper comments on an article by Schmidt and Hunter [Intelligence 27 (1999) 183.], who argue that the correction for attenuation should be routinely used in theory testing. It is maintained that Schmidt and Hunter's arguments are based on mistaken assumptions. We discuss our critique of Schmidt and Hunter in terms of two arguments against a routine use of the correction for attenuation within the classical test theory framework: (1) corrected correlations do not, as Schmidt and Hunter claim, provide correlations between constructs, and (2) corrections for measurement error should be made using modern test theory models instead of the classical model. The arguments that Schmidt and Hunter advance in favor of the correction for attenuation can be traced to an implicit identification of true scores with construct scores. First, we show that this identification confounds issues of validity and issues of reliability. Second, it is pointed out that equating true scores with construct scores is logically inconsistent with the classical test theory model itself. Third, it is argued that the classical model is not suited for detecting the dimensionality of test scores, which severely limits the interpretation of the corrected correlation coefficients. It is concluded that most measurement problems in psychology concern issues of validity, and that the correction for attenuation within classical test theory does not help in solving them.

In this article, grade point average (GPA) is considered a missing data technique for unavailable grades in school grade records. In Study 1, theoretical and empirical differences between GPA and seven alternative missing grade techniques were considered. These seven techniques are subject mean substitution, corrected subject mean, subject correlation substitution, regression imputation, expectation maximization algorithm imputation and two multiple imputation methods-stochastic regression imputation and data augmentation., The missing grade techniques differ greatly. Data augmentation and stochastic regression imputation appear to be superior as missing grade techniques. In Study 2, the completed grade records (observed and imputed values) were used in two prediction analyses of academic achievement. One analysis was based on unweighed grades, the other on weighed grades. In both analyses, alternative missing grade methods produced better and more consistent predictions. It is concluded that some alternative missing grade methods are superior to GPA.

The literature on thought experiments has been mainly concernedwith thought experiments that are directed at a theory, be it in aconstructive or a destructive manner. This has led somephilosophers to argue that all thought experiments can beformulated as arguments. The aim of this paper is to drawattention to a type of thought experiment that is not directed ata theory, but fulfills a specific function within a theory. Suchthought experiments are referred to as functional thoughtexperiments, and they are routinely used in applied statistics. An example is given from frequentist statistics, where a thoughtexperiment is required to establish the probability space. It isconcluded that (a) not all thought experiments can be formulated asarguments, and (b) the role of thought experiments is more generaland more important to scientific reasoning than has previouslybeen recognized.

A general definition of differential item functioning (DIF) is presented. The definition is specified for DIF between two groups and a test (questionnaire) that measures one latent trait. Methods for DIF detection are described for dichotomous, ordinal-polytomous, and continuous item responses under unidimensional item response theory (IRT) models. Keywords: measurement invariance; DIF definition, size and tests; unidirectional and bidirectional DIF; item response functions; test purification

Background: Animal studies demonstrated that early exposure to phenobarbital decreases reproductive function. This study investigates whether prenatal exposure to these anticonvulsants affects human genital tract development. Methods: Genital anomalies at birth were studied retrospectively in 90 phenobarbital-exposed, 108 phenobarbital plus phenytoin-exposed, and 198 matched control infants. Of this group, 72 drug-exposed males, 75 drug-exposed females, and 147 matched control subjects participated in a follow-up and were interviewed at age 19-35. Differences between groups were tested by chi-square and t-tests. Results: A total of 15% of the phenobarbital-exposed boys versus 2.8% control boys had undescended testes at birth. More anticonvulsant-exposed (24%) than control males (11%) had received medical treatment for genital anomalies. Anticonvulsant-exposed females more often had irregularities in menstrual cycles (31% vs 17%) and bleeding (15% vs 3%) and reported more problems during pregnancy. Conclusions: Prenatal exposure to anticonvulsants seems to induce minor genital anomalies and may affect reproductive function.

There is increasing pressure to effectively treat patients with complex care needs from the moment of admission to the general hospital. In this study, the authors developed a measurement strategy for hospital-based care complexity. The authors' four-factor model describes the interrelations between complexity indicators, highlighting differences between length of stay (LOS), objective complexity (such as medications or consultations), complexity ratings by the nurse, and complexity ratings by the doctor. Their findings illustrate illustrate limitations in the use of LOS as a sole indicator for care complexity. The authors show how objective and subjective complexity indicators can be used for early and valid detection of patients needing interdisciplinary care.

Animal studies have shown that prenatal exposure to the anticonvulsant drugs phenobarbital and phenytoin alters steroid hormone levels which consequently leads to disturbed sexual differentiation. In this study, possible sequelae of prenatal exposure to these anticonvulsants on gender development in humans were investigated. A follow-up study was carried out in phenobarbital- and phenytoin-exposed subjects and control subjects matched for age, sex, and the mothers' ages. Subjects were born in the Academic Medical Center between 1957 and 1972. Out of 243 exposed and 222 control subjects who were asked to volunteer, 147 exposed subjects (72 male, 75 female) and equal numbers of their matched control subjects participated in the follow-up study. They were interviewed and were asked to fill out questionnaires on gender role behavior, gender development, and sexual orientation. As a group, exposed and control subjects did not differ with respect to gender role behavior, although higher numbers of prenatally anticonvulsant-exposed subjects reported current or past cross-gender behavior and/or gender dysphoria. Three prenatally anticonvulsant-exposed subjects were transsexuals and had undergone sex reassignment surgery, a remarkably high rate given the rarity of transsexualism. In addition, two exposed males had exclusively homosexual experiences, whereas none of the control males reported exclusive homosexual behavior. The groups did not differ in attainment of pubertal psychosexual milestones.

This study aimed to assess whether previously established risk factors for sudden infant death syndrome (SIDS) are still valid now that the incidence in the Netherlands has dropped to 0.26per 1000 liveborn infants. A distinction was made between immutable and mutable risk factors. This case-control study (part of the European Concerted Action on SIDS) comprised 73 SIDS cases and 146 controls and lasted from March 1995 to September 1996. Adjustments were made for sleeping position and bedding factors by treating them as covariables. Apart from these factors, well known risk factors that remain of importance in the Netherlands are: male sex, young maternal age, twins, and low socioeconomic status. These factors are largely immutable. Other well known risk factors which might reflect attitudes to child care and could possibly be mutable are: smoking, alcohol consumption by the mother, bottle feeding, and change of babycare routine. Intervention strategies should focus on early signalling, thereby assisting parents in changing these unfavourable parenting attitudes. Information on optimal child care and extra support by public health nurses specifically aimed at families at risk could help to decrease further the incidence of SIDS in the Netherlands.

Because 'traditional' tracking analyses have some drawbacks, this paper presents a new method, which is based on generalized estimating equations (GEE). The new method is illustrated with data from the Amsterdam Growth and Health Study. In this observational longitudinal study six repeated measurements were carried out on 181 subjects (initial age 13 years) over a period of 15 years. Tracking was assessed for total cholesterol (TC), high density lipoprotein (HDL) and the TC/HDL ratio by calculating the odds ratio (OR) for subjects at risk at the age of 13 years regarding the development of their risk status over a 15 year period. These ORs can be interpreted as tracking coefficients. Three methods were compared: percentage of subjects who maintain their position in a certain risk group (i.e. univariate logistic regression), multivariate logistic regression and GEE. The three methods differ in the possibility of using all available data in the analysis and in the possibility of adjusting for certain covariates. Based on this, the GEE-approach seemed to be the most appropriate to calculate tracking coefficients for subjects at risk. When the risk groups were defined according to objective (absolute) risk values, for TC the GEE-OR was 10.1 (95% confidence interval (CI) 5.0-21.9), for HDL 14.4 (95% CI 7.2-28.7) and for the TC/HDL ratio 25.5 (95% CI 11.5-56.8). It can be concluded that the GEE-approach is very suitable to assess tracking for subjects at risk.

STUDY OBJECTIVES: To analyse the relation between the longitudinal development of total serum cholesterol (TC), high density lipoprotein cholesterol (HDL), and the TC/ HDL ratio and the longitudinal development of the biological parameters body fatness (SSF), lean body mass (LBM), and cardiopulmonary fitness (VO2-max). The relations were analysed with generalised estimating equations (GEE). SETTING: The relations were investigated with data from the Amsterdam growth and health study, a longitudinal study in which six measurements were carried out within a period of 15 years. PARTICIPANTS: Altogether 98 females and 84 males aged 13 years at the start of the study. MAIN RESULTS: Adjusted for lifestyle and other biological parameters, the longitudinal development of TC was inversely related to the development of LBM (standardised regression coefficient beta = -0.27; p < or = 0.01) and positively to SSF (beta males = 0.32; p < or = 0.01 and beta females = 0.15; p < or = 0.01). HDL was inversely related to LBM (beta = -0.26; p < or = 0.01) and positively to VO2-max (beta = 0.08; p < or = 0.05). The TC/HDL ratio was positively related to SSF (beta males = 0.39; p < or = 0.01 and beta females = 0.13; p < or = 0.01) and inversely to VO2-max (beta = -0.09; p < or = 0.05). CONCLUSIONS: The longitudinal analyses showed that body fatness was related to a high risk profile with respect to hypercholesterolaemia, and cardiopulmonary fitness to a low risk profile. Furthermore, it was shown that using body mass index as an indicator of body fatness in relation to lipoprotein values, has some important drawbacks.

In survey data, four potential sources of measurement error can jeopardize the results: the respondents, the interviewers, the questions, and the data collection method. In the past two decades, a shift has occurred in the way survey data are collected; telephone surveys and, to a lesser degree, mail surveys are now more extensively used. This has stimulated empirical research on the influence of the data collection method on data quality. Most of these mode comparisons used univariate criteria. In this study we concentrate on the potential influence of the data collection method on two substantive structural models. A controlled field experiment was conducted in which a mail, a telephone, and a face-to-face survey were compared. Using multigroup comparisons, two substantive structural equation models (one for loneliness and one for general well-being) were compared across the different data collection methods. The different data collection methods turned out to produce significantly different covariance matrices. Subsequent analyses showed that structural models, based on these covariance matrices, also differed.

The longitudinal relationship between serum levels of lipoproteins and lifestyle measures (e.g., intake of cholesterol, carbohydrates, saturated fatty acids [SFA], polyunsaturated fatty acids [PUFA], the PUFA/SFA (P:S) ratio, alcohol consumption, smoking behavior, and daily physical activity) was investigated over a period of 15 years in a cohort study (the Amsterdam Growth and Health Study). Members of the cohort were 13 years of age at commencement of the study and were 27 years old at termination. The unique feature of the study presented is that the longitudinal relations were analyzed with generalized estimating equations, a statistical technique in which the relations at different time-points are tested simultaneously. The development of the total serum cholesterol (TC) level was positively related to the intake of cholesterol and negatively to the P:S ratio. The development of high-density-lipoprotein cholesterol (HDL) level was positively related to alcohol consumption and daily physical activity and negatively to smoking behavior. The development of the TC:HDL ratio was negatively related to alcohol consumption and positively to both smoking behavior and carbohydrate intake.