Running head: NON-RESPONSE AND ATTRITION 1 1 2 3 An Analysis of Non-response and Attrition in the Zurich Project on Social 4 Development from Childhood to Adulthood (z-proso) 5 6 Nora Eisner1,2, Aja L. Murray3, Manuel Eisner3, Denis Ribeaud4 7 8 Submitted to International Journal of Behavioral Development 9 30 March 2018 10 Manuscript: 27 pages text, 5 tables, 1 figure 11 12 13 14 15 16 17 18 1 Contacting author: nora.eisner@hotmail.com 2 Department of Physics and Astronomy, University of Sheffield, Sheffield, UK 3 Institute of Criminology, University of Cambridge, Cambridge, UK 4 Jacobs Center for Productive Youth Development, University of Zurich, CH NON-RESPONSE AND ATTRITION 2 ABSTRACT 19 Selective non-participation and attrition pose a ubiquitous threat to the validity of 20 inferences drawn from observational longitudinal studies. We investigate various 21 potential predictors for non-response and attrition of parents as well as young persons at 22 different stages of a multi-informant study. Various phases of renewed consent from 23 parents and young persons allowed for a unique comparison of factors that drive 24 participation. The target sample consisted of 1675 children entering primary school at age 25 seven in 2004. Seven waves of interviews, over the course of ten years, measured levels 26 of problem behaviour as rated by children, parents and teachers. In the initial study 27 recruitment, where participation was driven by parental consent, non-response was 28 highest amongst certain socially disadvantaged immigrant minority groups. There were 29 fewer significant group differences at wave 5, when young people could be directly 30 recruited into the study. Similarly, attrition was higher for some immigrant background 31 groups. Methodological implications for future analyses are discussed. 32 33 34 35 Keywords: Criminology, Attrition, Non-response, Study-Participation, 36 Longitudinal Study 37 38 NON-RESPONSE AND ATTRITION - 3 - Longitudinal studies in normative samples are critical for making inferences about 39 the developmental processes underpinning child and adolescent development. Non-40 random participation and attrition, whereby presence in a research sample at a given wave 41 is directly or indirectly related to the characteristics under study, represent important 42 challenges for such studies (e.g. Singer & Willet, 2003). Knowledge of the nature and 43 severity of selective participation and drop-out is fundamental for applying appropriate 44 corrections to data analyses and can inform strategies to ameliorate selective participation 45 and drop-out in future studies and/or measurement waves. In this study, we therefore 46 sought to identify key parent and child characteristics that predict non-participation at 47 different stages in a 10-year multi-rater longitudinal study of youth development. 48 Child participation in longitudinal studies typically relies on parental consent until 49 the child is of an age where they are legally able to provide informed consent. It is also 50 common for parents to directly participate in the research as informants on their child’s 51 behaviour. As such, characteristics of both the parent and child can influence participation 52 and drop-out in studies of child and adolescent development. These characteristics are 53 often related to those under study (e.g. Audrain et al., 2002; Asendorpf et al., 2014; Noll 54 et al., 1997; Ullebo et al., 2012), resulting in the effects due to non-random participation 55 and non-random attrition undermining the representativeness of study samples. 56 Moreover, non-random drop-out can introduce spurious developmental effects or mask 57 genuine ones. Brame & Piquero (2003), for example, note that in longitudinal studies of 58 delinquent behaviour, those with the highest levels of delinquency tend to be more likely 59 to drop-out. In such cases, estimates of normative changes in delinquent behaviour over 60 development will be negatively biased because of the selective loss of higher scoring 61 participants. 62 NON-RESPONSE AND ATTRITION - 4 - To draw valid conclusions about developmental processes from longitudinal data, 63 relations between participation and study outcomes need to be taken into account. 64 Characterising non-random attrition has important implications for the interpretation of 65 statistical results from an affected dataset, as well as for the application of appropriate 66 mitigation strategies. Examples of such strategies include selection models, range 67 restriction corrections, data weighting, maximum-likelihood, multiple imputation, pattern 68 mixture models and random coefficient models (e.g. Asparouhov, 2005; Enders, 2011; 69 Sackett & Yang, 2000; Schafer & Graham, 2002). It is important to note that each method 70 comes with a different set of assumptions about the missing data mechanism. For 71 example, maximum-likelihood estimation for missing data yields unbiased parameter 72 estimates only if non-participation can be described as ‘missing at random’, that is, 73 participation is related to the observed but not the unobserved data (Rubin, 1976). When 74 participation is related to unobserved values over and above its relation to observed values 75 this is known as ‘missing not at random’ (MNAR). Here, methods such as pattern mixture 76 modelling or random coefficient modelling may be more appropriate; however, their 77 utility depends on how closely correlated drop-out and the variables of interest are (e.g. 78 Schafer & Graham, 2002). Even though in most cases there is insufficient information 79 about those who are missing to identify participation mechanisms empirically, analysing 80 patterns of drop-out can help researchers develop plausible hypotheses about these 81 mechanisms. These kinds of analyses also have the potential to inform future study 82 designs by providing a forecast of the profiles of individuals who may be most likely to 83 drop-out. In future waves or studies, special strategies may need to be developed and/or 84 additional resources channelled to their recruitment and retention (e.g. Eisner & Ribeaud, 85 2007). In this study, we therefore provide an analysis of participant factors predicting 86 NON-RESPONSE AND ATTRITION - 5 - non-participation and attrition in seven waves of a longitudinal study containing two 87 intervention arms: the Zurich Project on Social Development from Childhood to 88 Adulthood (z-proso). 89 Method 90 z-proso 91 Z-proso is an ongoing multi-rater longitudinal study of child and adolescent 92 development with a particular focus on the development of crime and aggression. The 93 study began in 2004 when the participants were entering their first year of school, aged 94 7. Parents provided data at four waves of interviews, when the children were aged 7, 8, 95 9, and 10 (labelled ‘waves P1 to P4’) and children provided self-report data when they 96 were aged 7, 8, 9 ,10, 11, 13, 15 and 17 (labelled ‘waves Y1 to Y7’). Additional 97 information was provided by teachers when the children were aged 7, 8, 9, 10, 11, 13 and 98 15 (labelled ‘waves T1 to T6’). 99 Sample 100 At baseline, a stratified random sampling approach was used to define the target 101 sample with schools as the randomisation units and stratification by school size and 102 socioeconomic background. The target sample comprised all children entering first grade 103 across 56 primary schools in the city of Zurich, Switzerland, corresponding to a total of 104 1,675 children. Lower socioeconomic neighbourhoods were slightly overrepresented. In 105 2004, when the study started, Zurich had a population of about 365,000, with a large 106 proportion of immigrant-background residents. Zurich is an affluent city. The average 107 GDP per capita was about USD 106,000 in 2004, and the unemployment rate was about 108 4%. Broadly representative of city demographics, the baseline target sample consisted of 109 39.3% German-speaking (mostly Swiss or German) primary caregivers. Over 60% of 110 NON-RESPONSE AND ATTRITION - 6 - primary caregiver were not native German speaking. The mean age of the target children 111 at entry into primary school was 6.85 years, and the average number of siblings was 1.15. 112 113 Recruitment 114 Considerable efforts were employed in order to maximise participation of the 115 baseline target sample, with a strong focus on recruiting caregivers with an immigrant 116 background who may be less likely to agree to participate in research studies. Recruitment 117 procedures are described in detail in Eisner & Ribeaud (2007). In brief, contact letters 118 were written in the ten most commonly spoken languages with native speakers of these 119 languages taking on the role of recruiting and interviewing participants. Monetary 120 incentives, translated support letters from school authorities, and the inclusion of 121 community stakeholders were also used to maximise participation. Bilingual information 122 packs with study information and consent forms (available in German, Albanian, 123 Portuguese, Serbian/Bosnian, Spanish, Tamil, Turkish, English, Croatian and Italian) 124 were sent to all non-German-speaking primary caregivers. Prospective participants who 125 did not respond to the initial information pack were contacted by phone. No upper limit 126 on the number of trial calls to be made was imposed, and in some cases more than 20 127 attempts were necessary before contact was possible. Parents who could not be reached 128 by phone were visited at home by a male and a female interviewer who explained the 129 study in more detail. To further encourage participation, shopping vouchers worth 20 130 CHF were offered to parents for their participation. At the beginning of the interviews, 131 parents were asked to sign an informed consent form for the participation of their child 132 as well as the participation of the child’s teacher. Non-respondents were re-contacted and 133 asked to consent to their child’s and his/her teacher’s participation only. Recruitment at 134 NON-RESPONSE AND ATTRITION - 7 - waves 2 and 3 followed similar procedures with the informed consent obtained from 135 parents at wave 1 covering the entire period from wave 1 to 3. Renewed consent, provided 136 by the parents, was required at wave 4. 137 Several changes were made to the recruitment and assessment procedures in wave 138 5. First, parent interviews were no longer carried out. Second, at age 13, youth were able 139 to actively consent to their own participation (Art. 16 of the Swiss Civil Code); however, 140 parents were still provided the opportunity to opt their child out. Third, unlike in previous 141 waves, the entire initial target sample defined at baseline could be re-contacted. Fourth, 142 youth interviews could no longer be carried out during regular school hours and thus 143 questionnaires were administered in classrooms outside of regular lesson times. In order 144 to maximise participation, participants received monetary incentives worth 30 CHF and 145 50 CHF in waves 5 and 6 respectively. Wave 6 required renewed active youth consent. 146 At this stage, the entire initial target sample could be re-contacted once again and the 147 monetary incentive increased to 60 CHF. The same recruitment procedure as described 148 above and preceding wave 5 was followed. 149 Given that the times of renewal of consent represented key attrition points, four 150 key outcomes can be defined with respect to non-random participation and attrition: 1) 151 baseline participation 2) participation in wave 5 3) drop-out in the wave 1 to 3 period 4) 152 drop-out in the wave 5 to 7 period. We analysed parent and child participation separately 153 given the previous evidence that patterns of participation may differ across these groups 154 (Asendorpf et al., 2014). 155 Measures and Statistical Procedure 156 We evaluated a range of predictors of attrition reflecting the core theoretical themes 157 of z-proso as well as additional possible risk factors for non-participation that could 158 NON-RESPONSE AND ATTRITION - 8 - inform recruitment and sampling design in future studies. For those who did not 159 participate at all, the only information available was on gender, primary caregiver 160 language, neighbourhood social class and neighbourhood familialism. 161 Small (special needs) Class. The school system in the city of Zurich differentiates 162 between regular and small classes. Small classes are intended to meet the specific needs 163 of children with difficulties such as developmental delays, behavioural problems, 164 language barriers, and/or learning difficulties. Typically, small classes have a size of ten 165 or less children, compared to around 20 in regular classes. The target sample comprised 166 9.1% of children attending a small class in year 1 of primary school. 167 Mother Tongue of Primary Caregiver. At the beginning of the study the City of 168 Zurich’s School Department provided the z-proso team with a contact database of all the 169 study participants and their parents. The most reliable proxy for the cultural background 170 of the primary caregiver was the mother tongue as, unlike nationality, this remains 171 unaffected by naturalization. Nonetheless, the mother tongue does not differentiate 172 between, for example, caregivers of German or Swiss or of Portuguese or Brazilian 173 background, and can thus be ambiguous to some extent. In the present analyses we 174 distinguish nine groups, namely German or Swiss German (39.3%), 175 Serbian/Bosnian/Croatian (10%), Albanian (9%), Portuguese (7%), Tamil (5.3%), Italian 176 (5.3%), Spanish (5.1%), Turkish (4.5%), and ‘other’ languages (14.5%). 177 Neighbourhood Social Class and Familialism. Neighbourhood characteristics 178 were derived from census and other data that was systematically collected by the City of 179 Zurich’s Office of Statistics. These data are aggregated at the level of the 212 statistical 180 zones. Neighbourhood social class was obtained as the factor score of three indicators, 181 namely the unemployment rate (2002), the (inverted) percentage of self-owned 182 NON-RESPONSE AND ATTRITION - 9 - households (2000), and the percentage of foreign nationals (2003). Similarly, mean 183 household size (2000), the percentage of residents aged below 20 years (2002), and 184 residential stability (measured as the percentage of households living in the same 185 statistical zone as five years earlier) were used as the three indicators in order to obtain 186 factor scores for neighbourhood familialism. The residential address of each study 187 participant was then assigned to its corresponding statistical zone and its related factor 188 scores. 189 Parent Education Level and Family Composition. Parental education was 190 measured using a dichotomous variable indexing whether the primary caregiver 191 possessed a university education or not. Single parent status was measured using a 192 dichotomous item measuring whether the second parent (usually the father) was living in 193 the same household as the primary caregiver. Based on this, 29.4% of households were 194 defined as single parent households. 195 Child Behaviour. Child behaviour was measured using the Social Behaviour 196 Questionnaire (SBQ; Tremblay et al., 1991) which captures prosocial behaviour, 197 aggression, anxiety, depression, and ADHD. The z-proso study team developed a multi-198 informant version with matched items across a parent, a child and a teacher version. The 199 parent questionnaire contained all questions (55 items) and was administered in computer 200 assisted personal interviewing (CAPI) home interviews, offered in ten different 201 languages. Responses were given on a 5-point Likert scale (‘never’ to ‘very often’). 202 Teachers were surveyed by mail. The teacher version used essentially the same 203 items and answer format as the parent version, with some adaptations to the school 204 context and some items from the full 55-item version omitted. 205 NON-RESPONSE AND ATTRITION - 10 - For children, a specially adapted self-administered multimedia version with 54 206 items was used in wave 1. Based on the concept of the “Dominique interactif” (Valla et 207 al., 2000), each behaviour was depicted in a drawing representing either a boy or a girl, 208 matched with the gender of the subject. A voice recorded on the laptop read out each item 209 that was worded in an age-adequate language. The child could then answer the question 210 by pushing a “yes” or a “no” button on the screen. 211 The SBQ distinguishes five major domains of children’s social behaviour and 212 scores were computed as averages of all items. Prosocial behaviour (e.g. ‘shows 213 sympathy to someone who has made a mistake’) was measured with ten items in the 214 parent (α = .77) version, seven items in the teacher version (α = .92) and ten items in the 215 child version (α = .59). Symptoms of Anxiety/Depression (e.g. ‘is too fearful or anxious’, 216 ‘has trouble enjoying him/herself’) were measured by nine items in the parent (α = .71) 217 version, by seven items in the teacher version (α = .90) and by nine items in the child 218 version (α = .62). Symptoms of attention deficit and hyperactivity measured by nine items 219 in the parent (α = .79) version, by 8 items in the teacher version (α = .94) and by eight 220 items in the child version (α = .58). The scale for non-aggressive conduct problems (e.g. 221 ‘steals at home’, ‘tells lies and cheats’) comprised nine items in the parent (α = .68) 222 version, six items in the teacher version (α = .81) and nine items in the child version (α = 223 .60). Aggressive behaviour was measured by 12 items in the parent (α =.79) version, by 224 11 items in the teacher version (α = .93) and by 12 items in the child version (α =.72). In 225 terms of item-level missingness, prior to the receipt of the dataset, within each set of SBQ 226 variables (e.g. all parent-reported items within a wave; all self-reported items within a 227 wave etc.), missing item scores had been singly imputed using an expectation-228 maximisation algorithm. The correlations between parent reports, youth self-reports, and 229 NON-RESPONSE AND ATTRITION - 11 - teacher reports are provided in Supplementary Table S1. These were modest at best 230 ranging from .04 (teacher- and parent- reports of internalising) up to .33 (teacher- and 231 parent- reports of ADHD). 232 Statistical Procedure 233 For descriptive purposes, simple logistic regressions with list-wise deletion were 234 used to evaluate the relations between predictors and non-participation or drop-out 235 without considering the effects of other predictors. Participation/retention was coded=0 236 and non-participation/drop-out was coded=1 such that odds ratios >1 reflect increased 237 likelihood of non-participation or drop-out. Specifically, the odds ratios reflect the ratio 238 of the odds of non-participation/drop-out at levels of the predictor separated by one unit. 239 For example, an odds ratio of 2 would indicate that the odds of dropping out double for 240 each unit increase in the predictor. Associated (unadjusted) p-values are reported for 241 descriptive purposes. These analyses were conducted in R statistical software, using a 242 logit link function in the glm function (R Core Team, 2016). 243 We then conducted a series of multiple regressions to evaluate the unique relations 244 between each predictor and drop-out/attrition controlling for other predictors. These 245 analyses were implemented in lavaan, again in R Statistical Software (Rosseel, 2012), 246 this time using probit regression. Probit and logistic regression can both be used to model 247 the prediction of dichotomous outcomes and generally result in the same conclusions. 248 While logistic regression uses a logit function to model the probability that the outcome 249 variable is equal to 1, probit regression uses an inverse standard normal cumulative 250 distribution function. Probit regression coefficients can thus be interpreted as the 251 difference in the cumulative normal probability of the outcome variable for a unit increase 252 in the predictor. Here, probit regression was used for practical reasons, namely that it (but 253 NON-RESPONSE AND ATTRITION - 12 - not logistic regression) can currently be combined with full information maximum 254 likelihood (FIML) estimation to account for missingness in lavaan. FIML provides 255 unbiased parameter estimates provided data are missing at random (MAR). Predictors 256 were entered using simultaneous entry. 257 To correct for multiple comparisons, we used the generalised Holm (1979) k-258 familywise error rate (FWER; Lehmano & Romano, 2005) method discussed by 259 Keselman et al. (2011). This method was selected because it is less conservative than 260 traditional FWER corrections which entail a substantial loss of statistical power when 261 families of statistical tests are large. While it is important to control type 1 errors, the 262 current study is somewhat exploratory in nature, and it was judged more problematic to 263 fail to identify predictors that were associated with attrition than to falsely conclude that 264 a predictor was related to attrition when it is not. This viewpoint, in turn, comes from the 265 importance of recognising when and in what way attrition is non-random in order to guard 266 against biases deriving from falsely assuming that attrition is random. 267 In the Holm k-FWER method, k-FWER is defined as the probability of rejecting at 268 least k hypotheses Hi where i is a member of the set of true null hypotheses. Thus when 269 k is one, this reduces to the traditional FWER correction i.e. that which controls the 270 probability of rejecting at least one true null hypothesis. 2-FWER controls the probability 271 of rejecting 2 or more true null hypotheses (implicitly tolerating one false positive), 3-272 FWER, the probability of rejecting 3 or more true null hypotheses and so on and so forth. 273 Keselman et al. (2011) recommend selecting a value for k that is the nearest integer to mα 274 where m is the number of tests conducted and α is the significance level. In total, 129 tests 275 were conducted, giving a k value of 6. This was judged an acceptable level given the goals 276 of the study. 277 NON-RESPONSE AND ATTRITION - 13 - The method involves considering the raw p-values for all the hypotheses, ordered 278 from smallest to largest and making a sequential adjustment to the alpha value for each 279 or, equivalently, to the p-value itself. We here report adjusted p-values rather than 280 adjusted critical values on the assumption that the former will be more informative for 281 most readers. Both unadjusted and adjusted p-values are reported in Table 2 for 282 information; however, all inferences were made on the basis of adjusted p-values. 283 Results 284 Descriptive Statistics 285 Figure 1 shows the number of participating primary caregivers (subscript P) and 286 youth (subscript Y) in each of the main data-collection waves. In wave 1, 1239 primary 287 caregivers (74% of baseline target sample) participated and a further 121 parents provided 288 consent for participation of the child only, meaning that 1360 young persons participated 289 in wave 1. Between waves 1 and 2, attrition was low with 4.5% and 2.0% of primary 290 caregivers and youth dropping out respectively. 291 The bold arrows in Figure 1 show the pool of participants that could be re-contacted 292 in each wave of the study, whilst the thin arrows highlight the number of subjects lost due 293 to attrition between consecutive waves. In wave 4, for example, all participants that 294 consented to participate in wave 1 could be re-contacted, resulting in a parent 295 participation of NP= 1075 (64% of baseline target sample) and young person participation 296 of NY = 1147 (69%). The number of parents and youth lost due to attrition between waves 297 3 and 4 was NP=116 (9.8% of the parents in Wave 3) and NY= 184 (14%). Unlike in the 298 initial recruitment into wave 1, where 121 youth participated even though their parents 299 did not, the number of ‘youth-only’ cases were small in wave 4. The total number of 300 parents and youth who re-entered into wave 4 were NP= 11 and NY = 10 respectively. 301 NON-RESPONSE AND ATTRITION - 14 - These participants missed one or more waves of data collection but subsequently 302 continued to participate. This is depicted by dashed lines in figure 1. The lowest 303 participation rate occurred following the request for renewed parental consent in wave 4. 304 The highest number of participation followed in wave 6, where the whole initial target 305 sample was re-contacted, with NY = 1446 (86% of baseline target sample). Finally, after 306 wave 7, one participant requested to be withdrawn from the study and have their data 307 removed from the database. This participant is represented in subsequent analyses as if 308 they did not participate at baseline and remained a non-participant thereafter. Omitting 309 this participant, N=1570, or 94% of the target sample participated in at least one of the 310 seven waves. 311 Means/frequencies for each predictor of attrition, broken down by attrition status, 312 are provided in Tables 1-3. These tables also provide simple logistic regression analyses 313 to test the unadjusted effects of each predictor prior to controlling for other predictors. 314 The sample sizes in the table also illustrate the level of variable-wise missingness at each 315 wave. 316 Non-participation 317 Results of the multiple probit regression models assessing predictors of parent and 318 youth non-participation in wave 1 (average age 7) and wave 5 (average age 13) are 319 provided in Table 4. After correction for multiple comparisons, parent non-participation 320 in wave 1 was significantly predicted by neighbourhood social class (b=-0.05) and several 321 non-German first languages (ranging in effect from b=0.18 for Portuguese to b=0.30 for 322 Tamil and Albanian). Youth non-participation in wave 1 was significantly predicted by 323 the same non-German caregiver first language categories, except Tamil (ranging in effect 324 from b=0.17 for Portuguese, Serbian-Croatian and Albanian up to b= 0.27 for Turkish). 325 NON-RESPONSE AND ATTRITION - 15 - The only predictor with a significant unique effect on youth participation at wave 5 was 326 membership in a small class, which was associated with a greater probability of non-327 participation (b=0.12). 328 Results of the multiple probit regression models assessing predictors of dropping out 329 are provided in Table 5. Although there were many significant effects on drop-out when 330 considering bivariate models, there were few predictors that had significant unique effects 331 on attrition in the multiple regression models after correcting for multiple comparisons. 332 Being in the ‘Other’ language category significantly predicted both parent (b=0.13) and 333 youth (b= 0.10) drop-out between waves 1 and 4, while being in the ‘Serbian-Croatian’ 334 category predicted youth drop-out between waves 5 and 7 (b=0.10). As there were no 335 strong a priori theoretical rationale for expecting higher-order effects, including 336 interactions between predictors, these were not considered for any of the models. 337 Discussion 338 Longitudinal studies inevitably suffer from survey non-response and attrition. This 339 not only results in a smaller dataset, thus reducing the power of the study, but also has the 340 potential of introducing bias. We evaluated a number of different predictors of non-341 response and attrition, including characteristics of parents and children as rated by 342 parents, children and teachers. 343 We found that child and parent non-participation and drop-out were more likely 344 among children with primary caregivers who spoke languages other than the official 345 regional language. This was in spite of specific and intensive efforts to encourage 346 participation among non-German speaking, immigrant background parents (e.g. Eisner & 347 Ribeaud, 2007). With the exception of Italian-speaking respondents, over 90% of non-348 German speaking households had at least one first generation immigrant parent. Thus, 349 NON-RESPONSE AND ATTRITION - 16 - non-native speaking primary caregivers is likely to be a proxy for belonging to an 350 immigrant minority. In this respect, our results are consistent with those from other 351 European survey-based studies that suggest that immigrant minorities are generally more 352 difficult to contact and are more likely to decline to participate (Couper & Leeuw, 2003; 353 van Goor et al., 2005; Kapteyn et al., 2006). 354 Even though we were unable to collect data on reasons for non-participation among 355 these individuals, it is possible to speculate as to its cause. Some participants may have 356 felt intimidated by the prospect of participating in an interview in a foreign country. 357 Furthermore, sociocultural minorities are often in a more vulnerable position. Amongst 358 other factors, their legal residence status may still be undecided, they may have had 359 adverse experiences with immigration authorities or they may have faced persecution in 360 their home countries, which could generalise to a lack of trust in research studies. This 361 may be exacerbated by study topics that could be considered sensitive. Third, immigrant 362 minority caregivers have, on average, a more limited educational background and lower 363 socio-economic status (Eisner et al., 2007). Multiple studies have shown that educational 364 level is an important predictor of participation in research (e.g. Stoop, 2005; Van Loon et 365 al., 2003; Korkeila et al., 2001; Curtin, Presser & Singer, 2000). However, in this study 366 whether or not caregivers had a university education was not a significant predictor of 367 parent or youth drop-out over the early waves. However, we did not have parental 368 education data for parents who did not participate in the study at all, meaning that we 369 could only indirectly test whether this was associated with participation at wave 1 based 370 on FIML estimates. 371 Finally, cultural factors may have played a role. Whilst in many countries it is 372 common custom to participate in surveys and answer personal questions, in other cultures 373 NON-RESPONSE AND ATTRITION - 17 - giving personal information to a stranger is counter-normative (e.g., Johnson et al., 2002). 374 Though we have no direct evidence that this was the case in z-proso, this could be one 375 aspect to explore in future research. Overall, our results suggest that translation of 376 invitation letters and additional efforts to contact immigrant background participants may 377 be insufficient to mitigate the tendency to decline to participate and to drop out. Differing 378 cultural attitudes and sources of apprehension must also be addressed as part of 379 recruitment strategies. That said, there is some indirect evidence that the additional efforts 380 invested in recruiting individuals from an immigrant minority status helped mitigate 381 under-representation of these individuals in the sample. For example, at baseline, a higher 382 proportion of those from an immigrant minority background were recruited via more 383 ‘active’ methods (active telephone recruitment) than more ‘passive’ methods 384 (participants answering an invitation by response slip; Eisner & Ribeaud, 2007). 385 Specifically, 28% of the target sample from non-German speaking minorities were 386 recruited via the return of a response slip, while 33.7% were recruited by telephone. In 387 contrast, among the target sample of German-speakers 63.8% were recruited by reply slip 388 and 24% via telephone contact. As home visit recruitments were extremely rare, they 389 were collapsed with the telephone recruitment category. This suggests that investing 390 additional efforts in recruiting immigrant minorities can improve response rates and that, 391 in fact, these additional efforts may produce greater returns in immigrant minority groups 392 than in non-minority groups. 393 Although they had no unique significant effect on non-participation, itt is also 394 worth highlighting the bivariate association between child behaviour and non-395 participation as an area for potential future investigation. Based on bivariate analyses and 396 according and teacher reports of child behaviour, child aggression, non-aggressive 397 NON-RESPONSE AND ATTRITION - 18 - conduct problems, ADHD, internalising and prosociality predicted parental and youth 398 drop-out during the wave 1 to 4 phase. Specifically, both parents and youth were more 399 likely to drop-out if the child showed higher levels of psychopathology, whereas parents 400 reported a greater likelihood of drop-out where their child showed lower levels of 401 psychopathology. These results are consistent with past research, which suggests that 402 children exhibiting disruptive behaviour often show difficulties functioning within the 403 school context (e.g. Barry et al., 2002). Moreover, in terms of participation in research 404 studies, by definition children with high levels of ADHD symptomology may avoid and 405 find it very difficult to engage in tasks that include sustained attention (APA, 2013). This 406 will almost certainly include completing study measures. Further, it has been suggested 407 that parents of children with conduct problems may have a tendency to ‘disengage’ from 408 the child and their problem behaviour; an effect which would presumably carry over into 409 participation in studies (e.g. Patterson, 1982). As such, studies seeking to retain the most 410 disruptive children must take into account possible difficulties such as the child having a 411 poor bond or negative associations with school context, difficulties completing measures 412 and parents who are avoidant of or psychologically disengaged from their child’s 413 behaviour. In these cases, measures such as offering assessments outside of the school 414 setting and breaking up the assessment into multiple sessions may help mitigate the loss 415 of participants with ADHD and/or disruptive behaviours. 416 Notably, despite positive correlations between parent and teacher reports on all 417 behaviours, parent reports of child behaviour suggested exactly the opposite pattern to 418 that suggested by teachers. Parent-reports suggested that attrition was more likely when 419 their child was well-adjusted. While it is common for different raters to disagree on child 420 behaviour (e.g. De Los Reyes, 2011), that these two raters suggested completely 421 NON-RESPONSE AND ATTRITION - 19 - contradictory results is striking. One possibility is that children show context-specific 422 negative behaviour and that negative behaviour in the school environment is particularly 423 undermining of tendencies to engage with research projects conducted in the school 424 setting. Another possibility, however, is that socially desirable responding on the part of 425 parents with children displaying disruptive behaviours accounts for the counterintuitive 426 association between parent-reported child behaviour and study participation (e.g. Johnson 427 et al., 2012; Mundia, 2011; Eisner & Ribeaud, 2007). However, it is also important to 428 emphasise that in the multiple regressions, neither teacher nor parent-reported child 429 behaviour was uniquely associated was attrition. Thus, any effects of behaviour cannot 430 be easily disentangled from one another nor from the effects of other factors that may 431 affect non-participation, such as immigrant minority status discussed above. 432 Our result suggested that the dynamics of attrition and non-response changed once 433 children/youths could provide their own active consent, highlighting the need for different 434 retention methods based on the target sample. We found that being able to re-contact the 435 entire initial target sample going into waves 5 and 6 greatly improved our overall 436 participation, resulting in a highest young person participation rate occurring in wave 6. 437 We note that re-contacting entire target samples is not always a viable option, and 438 regulations may vary depending on location, nonetheless, we emphasise the positive 439 outcomes this can have on participation rates. Furthermore, it is important to identify 440 more vulnerable samples in the early stages of the study, so that long term retention plans 441 can be implemented in order to retain or regain participants. 442 Collectively, our results suggested that participation and attrition in z-proso are 443 related to characteristics of the children under study. As such, our data would not qualify 444 as missing completely at random (MCAR) rendering analysis methods involving listwise 445 NON-RESPONSE AND ATTRITION - 20 - or pairwise deletion inappropriate. Under the assumption of MAR, methods such as 446 maximum likelihood estimation or multiple imputation could likely address parameter 447 bias due to non-random attrition. The question of whether the data qualifies as MAR is 448 more difficult to answer because this would require knowledge of unobserved data. 449 However, given the comprehensiveness of the list of measures obtained in z-proso (see 450 Ribeaud & Eisner, 2010) and the relatively low rates of attrition observed overall, we 451 would argue that, provided a sufficient set of auxiliary variables is used, any additional 452 attrition following a NMAR mechanism may have only a small effect. We, therefore, 453 believe that users of z-proso dataset could be justified in employing methods such as 454 multiple imputation or maximum likelihood estimation drawing on a range of auxiliary 455 variables for testing developmental hypotheses. However, it is always best practice to use 456 a range of methods, especially those that make different assumptions about mechanisms 457 of missingness (e.g. pattern mixture modelling versus FIML estimation) to evaluate the 458 sensitivity of results to the chosen method of dealing with missing data. In addition, our 459 goal here was not to provide an exhaustive analysis of all the factors that may predict 460 attrition for the purposes of aiming to achieve MAR but to focus on a smaller number of 461 theoretically motivated predictors that may help researchers to understand why certain 462 individuals may be more likely to drop-out than others. This in turn can help inform 463 recruitment and retention in future studies and future waves of existing longitudinal 464 studies. Thus, we would advise users of the dataset to consider an even broader range of 465 predictors as auxiliary variables, as well as their non-linear effects and interactions. 466 Finally, it is important to consider the limitations of the current study. First, we did not 467 ask participants to report on their reasons for non-response. This would have allowed us 468 to understand proximal causes of non-participation. Second, we had only limited 469 NON-RESPONSE AND ATTRITION - 21 - information on those who declined to participate at baseline. Third, we included a limited 470 number of predictors of attrition, focussing specifically on those that represented the core 471 themes of z-proso and/or which were predicted to show a relation to attrition based on 472 past research. Future research will be valuable in identifying other variables that may be 473 related to selective non-participation. 474 Conclusion 475 Studying patterns of attrition has the potential to inform future study design, as well 476 as provide information pertinent to the interpretation or correction of statistical effects in 477 studies affected by non-random participation. In this study, we evaluated the 478 characteristics associated with non-participation and attrition in a 10-year longitudinal 479 study of child development. The design of the study allowed for a unique comparison of 480 factors associated with non-participation of parents and youths as well as attrition due to 481 various phases of renewed consent. This is because during the initial waves of the study 482 the decisions on study participation were made by the primary caregiver of the target 483 subject, whereas in later waves the youths themselves were able to provide informed 484 consent. We found several predictors of non-participation and attrition, including 485 characteristics of both the parent and child. 486 Our results showed that non-response and attrition was highest among certain 487 immigrant background, non-native speaking parents, despite the significant efforts made 488 to recruit these subjects. This result is in agreement with previous findings (e.g. Couper 489 & Leeuw, 2003) and could be due to a number of reasons, reviewed above. 490 Child behaviour was also found to be a significant predictor of attrition in the bi-491 variate analyses. Between waves 1 and 4, aggression, non-aggressive conduct problems, 492 ADHD, internalising and prosociality, as rated by teachers were all predictors of child 493 NON-RESPONSE AND ATTRITION - 22 - drop-out. This was in direct disagreement with the reporting of parents, which suggested 494 that children with lower levels of problem behaviours were less likely to participate. A 495 possible explanation for these contradictory results relates to social desirability bias, 496 where parents of children with disruptive behaviour are more likely to project a good 497 image of themselves by responding in the ‘most desirable’ way. A further explanation 498 could be context specific behaviour of the child, resulting in a disagreement between 499 parents and teachers. 500 Acknowledgements 501 We are grateful to the children, parents and teachers who provided data for the z-502 proso study and the research assistants involved in its collection. Funding from the Jacobs 503 Foundation (Grant 2010-888) and the Swiss National Science Foundation (Grants 504 100013_116829 & 100014_132124) is also gratefully acknowledged.505 NON-RESPONSE AND ATTRITION 23 References American Psychiatric Association, APA (2013). Diagnostic and statistical manual of mental disorders (DSM-5). Arlington, VA: APA. Audrain, J., Tercyak, K.P., Goldman, P., & Bush, A. (2002). Recruiting adolescents into genetic studies of smoking behavior. Cancer Epidemiology, Biomarkers & Prevention, 11(3), 249-252. Asendorpf, J. B., Van De Schoot, R., Denissen, J. J., & Hutteman, R. (2014). Reducing bias due to systematic attrition in longitudinal studies: The benefits of multiple imputation. International Journal of Behavioral Development, 38(5), 453-460. Asparouhov, T. (2005). Sampling weights in latent variable modeling. Structural Equation Modeling, 12(3), 411–434. Barry, T. D., Lyman, R. D., & Klinger, L. G. (2002). Academic underachievement and attention-deficit/hyperactivity disorder: The negative impact of symptom severity on school performance. Journal of School Psychology, 40(3), 259-283. Brame, R., & Piquero, A. R. (2003). Selective attrition and the age-crime relationship. Journal of Quantitative Criminology, 19(2), 107-127. Couper, M. P., & Leeuw, E. D. d. (2003). Nonresponse in cross-cultural and cross national surveys. In J.A. Harkness, F. J. R. van de Vijver and P. P. Mohler (eds.) Cross-Cultural Survey Methods, 157–178. New York: Wiley. Curtin, R., Presser, S., & Singer, E. (2000). The effects of response rate changes on the index of consumer sentiment. Public Opinion Quarterly, 64(4), 413-428. De Los Reyes, A. (2011). Introduction to the special section: More than measurement error: Discovering meaning behind informant discrepancies in clinical assessments of children and adolescents. Journal of Clinical Child and Adolescent Psychology, 40(1), 1–9. NON-RESPONSE AND ATTRITION 24 Eisner, M., & and Denis, R. (2007). Conducting a criminological survey in a culturally diverse context: Lessons from the Zurich Project on the Social Development of Children. European Journal of Criminology, 4(3), 271-298. Enders, C. K. (2011). Analyzing longitudinal data with missing values. Rehabilitation Psychology, 56(4), 267-288. Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian journal of statistics, 65-70. Johnson, T. P., Fenrich, M., & Mackesy-Amiti M. E. (2012). An evaluation of the validity of the Crowne–Marlowe need for approval scale. Quality & Quantity, 46(6), 1883-1986. Johnson, T. P., O’Rourke, D., Burris, J., & Owens, L. (2002). Culture and survey nonresponse. In R. M. Groves, D. A. Dillman, J. L. Eltinge and R. J. A. Little (eds) Survey Nonresponse, 55–69. New York: Wiley. Kapteyn, A., Michaud, P-C., Smith, J., & van Soest, A. (2006). Effects of attrition and non- response in the Health and Retirement Study. Discussion paper from the Institute for the Study of Labor (IZA), Bonn, Germany. Keselman, H. J., Miller, C. W., & Holland, B. (2011). Many tests of significance: New methods for controlling type I errors. Psychological Methods, 16(4), 420-431. Korkeila, K., Suominen, S., Ahvenainen, J., Ojanlatva, A., Rautava, P., Helenius, H., & Koskenvuo, M. (2001). Non-response and related factors in a nation-wide health survey. European Journal of Epidemiology, 17(11), 991-999. Lehmann, E. L., & Romano, J. P. (2012). Generalizations of the familywise error rate. In Selected Works of EL Lehmann(pp. 719-735). Springer, Boston, MA. Mundia, L. (2011). Social desirability, non-response bias and reliability in a long self-report measure: illustrations from the MMPI-2 administered to Brunei student teachers. Educational Psychology, 31(2), 207-224. NON-RESPONSE AND ATTRITION 25 Noll, B., Zeller, M.H., Vannatta, K., Bukowski, W.M., & Davies, W.H. (1997). Potential Bias in classroom research: Comparison of children with permission and those who do not receive permission to participate. Journal of Clinical Child Psychology, 26(1), 36-42. Patterson, G. R. (1982). Coercive family process: A social learning approach (Vol. 3). Eugene, OR: Castalia. R Core Team (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. Ribeaud, D., & Eisner, M. (2010). Risk factors for aggression in pre-adolescence: Risk domains, cumulative risk and gender differences-Results from a prospective longitudinal study in a multi-ethnic urban sample. European Journal of Criminology, 7(6), 460-498. Rosseel, Y. (2012). Lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 1-36. URL http://www.jstatsoft.org/v48/i02/. Rubin, D. B. (1976). Inference and missing data. Biometrika, 63(3), 581–592. Sackett, P. R., & Yang, H. (2000). Correction for range restriction: An expanded typology. Journal of Applied Psychology, 85(1), 112–118. Schafer, J.L. & Graham, J.W (2002). Missing data: our view of the state of the art. Psychological Methods, 7(2), 147–177. Singer, J. D., & Willet, J. B. (2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York: Oxford University Press. Stoop, I. A. L. (2005). The hunt for the last respondent. Nonresponse in sample surveys. The Hague: Social and Cultural Planning Office of the Netherlands. Tremblay, R. E., Loeber, R., Gagnon, C., Charlebois, P., Larivee, S., & LeBlanc, M. (1991). Disruptive boys with stable and unstable high fighting behavior patterns during junior elementary school. Journal of Abnormal Child Psychology, 19(3), 285-300. NON-RESPONSE AND ATTRITION 26 Ullebø, A. K., Posserud, M. B., Heiervang, E., Obel, C., & Gillberg, C. (2012). Prevalence of the ADHD phenotype in 7-to 9-year-old children: effects of informant, gender and non- participation. Social Psychiatry and Psychiatric Epidemiology, 47(5), 763–769. Valla, J.P., L. Bergeron, M. St-Georges et C. Berthiaume (2000). Le Dominique Interactif: présentation, cadre conceptuel, propriétés psychométriques, limites, et utilisations. Revue Canadienne de Psycho-éducation, 29(2), 327-347. Van Goor, H., Jamsma, F. and Veenstra, R. (2005). Differences in undercoverage and nonresponse between city neighbourhoods in a telephone survey. Psychological Reports, 96(3), 867–878. Van Loon, A.J.M., Tijhuis, M., Picavet., H.S.J., Surtees P.G. and Ormel, J. (2003) Survey non- response in the Netherlands: Effects on prevalence estimates and associations. Annals of Epidemiology, 13(2), 105–110. NON-RESPONSE AND ATTRITION 27 Figures Figure 1: Flow of subjects across seven waves of interviews. Highlighting the rate of attrition between consecutive waves as well as the number of subjects re-entering the study from previous waves. Wave 6 Young Persons (age 15) = 1446 (86%) Wave 7 Young Persons (age 17) = 1305 (78%) Active Young Person Consent for W7 Attrition Y = 36 (2.6%) Attrition Y = 159 (11%) Wave 5 Young Persons (age 13) = 1365 (81%) Re-enter Y = 117 Re-enter Y = 18 Phase 1 Active Parent Consent for W1-W3 Wave 3 Parents = 1180 (71%) Young Persons (age 9) = 1321 (79%) Attrition P = 56 (4.5%) Y = 36 (2.6%) Attrition P = 116 (9.8%) Y = 184 (14%) Wave 4 Parents = 1075 (64%) Young Persons (age 11) = 1147 (69%) Active Parent Consent for W4 Phase 2 Active Young Person Consent and Passive Parent Consent for W5- W6 Attrition P = 24 (2.0%) Y = 19 (1.4%) Re-enter P = 13 Y = 4 Re-enter P = 6 Y = 5 Re-enter P = 5 Y = 2 Wave 2 Parents = 1192 (71%) Young Persons (age 8) = 1334 (80%) Wave 1 Parents = 1239 (74%) Young Persons (age 7) = 1360 (81%) Target Sample P,Y = 1675 Re-enter P = 8 Y= 10 Re-enter P = 0 Y = 2 Re-enter P = 0 Y = 3 NON-RESPONSE AND ATTRITION 28 Table 1: Means/frequencies of participant background predictors and bivariate associations (ORs) with baseline response status Note. aReference category. *Significant at alpha<.05. N = 1665 – 1675. Predictor Non-Response P1 Non-Response Y1 Non-Response Y5 Participants Non- participan ts OR p Participants Non- participant s OR p Participants Non- participant s OR p Class size Normal 1139 378 1.70 <.01* 1244 273 1.60 .02* 1254 263 1.96 <.001 Small 101 57 117 41 112 46 Gender Male 643 226 1.01 .92 696 174 0.84 .17 703 167 0.90 .41 Female 596 209 665 140 663 142 Caregiver higher education - - - - - - - - - - - - Neighbourhoo d Social Class -0.34 -0.67 0.65 <.01* -0.37 -0.67 0.67 <.001 * -0.42 -0.47 0.93 .33 Neighbourhoo d Familialism 0.54 0.51 0.96 .51 0.53 0.56 1.03 .61 0.52 0.60 1.08 .20 Language Albanian 80 70 5.07 <.001 * 105 45 3.55 <.001 * 120 30 1.32 .23 Other 168 39 1.34 .16 184 23 1.04 .89 158 49 1.64 .01* Germana 562 97 - - 588 71 - - 554 105 - - English 23 3 0.76 .65 23 3 1.08 .90 22 4 0.96 .94 Italian 66 22 1.93 .01* 73 15 1.70 .09 69 19 1.45 .18 Portuguese 78 40 2.97 <.001 * 84 34 3.35 <.001 * 100 18 0.95 .85 Serbian-Croatian 105 62 3.42 <.001 * 117 50 3.54 <.001 * 138 29 1.11 .65 Spanish 63 22 2.02 .01* 70 15 1.77 .07 59 26 2.33 <.001 * Turkish 44 32 4.21 <.001 * 46 30 5.40 <.001 * 61 15 1.30 .40 Tamil 48 41 4.95 <.001 * 67 22 2.72 <.001 * 77 12 0.82 .55 NON-RESPONSE AND ATTRITION 29 Table 2: Means/frequencies of participant background predictors and bivariate associations (ORs) with attrition status Note. HE= higher education. aReference category. *Significant at alpha<0.05 Predictor Attrition P1 – P4 (N = 1240) Attrition Y1 – Y4 (N = 1217 - 1361) Attrition Y5 – Y7 (N = 1054 - 1366) Retained Dropped out OR p Retained Dropped out OR p Retained Dropped out OR p Class Size Normal 982 157 1.09 .77 1046 198 1.66 .03* 1119 135 1.38 .26 Small 86 15 89 28 96 16 Gender Male 556 89 1.01 .94 578 118 0.95 .73 610 93 0.63 .01* Female 512 83 557 108 605 58 Caregiver Higher Education Higher Education 780 136 0.72 .10 761 136 0.75 .17 689 90 0.31 <.001* No Higher Education 288 36 282 38 264 11 Neighbourhood Social Class -0.32 -0.49 0.82 .03* -0.35 -0.52 0.81 .01* -0.38 -0.62 0.73 .01* Neighbourhood Familialism 0.54 0.57 1.03 .74 0.52 0.58 1.06 .41 0.54 0.54 1.01 .87 Language Albanian 60 20 3.75 <.001 * 79 26 2.56 <.001 * 96 24 3.24 <.001 * Other 131 40 3.40 <.001 * 140 48 2.52 <.001 * 151 15 1.23 .52 Germana 516 46 - - 521 67 - - 515 39 - - English 20 3 1.69 .41 22 1 0.35 .32 19 3 2.13 .24 Italian 60 6 1.12 .80 65 8 0.96 .92 61 8 1.77 .16 Portuguese 63 15 2.63 <.001 * 71 13 1.42 .27 82 18 3.04 <.001 * Serbian-Croatian 84 21 2.81 <.001 * 86 31 2.80 <.001 * 110 28 3.45 <.001 * Spanish 51 12 2.59 .01* 54 16 2.30 .01* 54 5 1.11 .84 Turkish 37 7 2.13 .09 38 8 1.64 .22 54 7 1.74 .20 Tamil 46 2 0.45 .28 59 8 1.04 .92 73 4 0.75 .59 NON-RESPONSE AND ATTRITION 30 Table 3: Means of participant behavioural predictors and bivariate associations (ORs) with attrition status Note. *Significant at alpha<.05. Predictor Attrition P1 – P4 Attrition Y1 – Y4 Attrition Y5 – Y7 Retained Dropped out OR p Retained Dropped out OR p Retained Dropped out OR P Parent reports (NP1-P4 = 1230, NY1-Y4 = 1207, NY5-Y7 = 1044) Prosocial Behaviour 2.56 2.66 1.43 .02* 2.57 2.58 1.04 .79 2.57 2.59 1.07 .73 Internalising 0.72 0.61 0.59 .01* 0.71 0.67 0.80 .22 0.71 0.64 0.68 .11 ADHD 1.23 1.13 0.78 .07 1.21 1.22 1.01 .93 1.22 1.16 0.86 .37 Non-aggressive Conduct Disorder 0.61 0.48 0.38 <.001 * 0.60 0.55 0.68 .08 0.60 0.53 0.59 .07 Aggression 0.62 0.51 0.53 <.001 * 0.61 0.57 0.78 .21 0.61 0.55 0.70 .17 Self-reports (NP1-P4 = 1215, NY1-Y4 = 1359, NY5-Y7 = 1156) Prosocial Behaviour 0.82 0.82 0.91 .84 0.82 0.80 0.53 .11 0.82 0.82 1.06 .92 Internalising 0.41 0.43 1.32 .44 0.41 0.42 1.29 .41 0.41 0.41 0.89 .78 ADHD 0.17 0.17 0.95 .91 0.17 0.19 1.84 .10 0.17 0.17 0.79 .66 Non-aggressive Conduct Disorder 0.21 0.22 1.14 .77 0.21 0.23 1.79 .13 0.22 0.22 1.29 .63 Aggression 0.18 0.16 0.63 .37 0.17 0.18 1.10 .82 0.17 0.17 0.77 .66 Teacher reports (NP1-P4 = 1214, NY1-Y4 = 1338, NY5-Y7 = 1144) Prosocial Behaviour 2.20 2.13 0.90 0.3 2.19 2.05 0.81 .02* 2.20 2.06 0.81 .08 Internalising 0.85 0.90 1.09 .44 0.86 0.97 1.21 .04* 0.85 0.90 1.09 .50 ADHD 1.20 1.48 1.30 <.001 * 1.20 1.45 1.27 <.001 * 1.22 1.30 1.09 .39 Non-aggressive Conduct Disorder 0.30 0.43 1.58 <.001 * 0.30 0.45 1.74 <.001 * 0.31 0.36 1.25 .25 Aggression 0.55 2.13 1.39 <.001 * 0.56 0.75 1.45 <.001 * 0.57 0.62 1.12 .41 NON-RESPONSE AND ATTRITION 31 NON-RESPONSE AND ATTRITION 32 Table 4: Multiple probit regression results predicting non-response by participant background Predictor Outcome= Non-Response P1 Outcome= Non-Response Y1 Outcome= Non-Response Y5 b SE p 6-FWER Holm corrected p b SE p 6-FWER Holm corrected p b SE p 6-FWER Holm corrected p Small Class (reference category = small class) 0.04 0.04 .31 1.00 0.03 0.03 .36 1.00 0.12 0.03 <.001* .01* Gender (reference category = female) 0.01 0.02 .78 1.00 -0.02 0.02 .20 1.00 -0.01 0.02 .55 1.00 Caregiver Higher Education 0.10 0.05 .03 .60 -0.02 0.04 .60 1.00 0.01 0.03 .70 1.00 Neighbourhood Social Class -0.05 0.01 <.001* .02* -0.02 0.01 .20 1.00 0.00 0.01 .96 1.00 Neighbourhood Familialism 0.00 0.01 .67 1.00 0.00 0.01 .67 1.00 0.01 0.01 .24 1.00 Language (reference category = German) Albanian 0.30 0.04 <.001* <.001* 0.17 0.04 <.001* <.001* 0.03 0.04 .34 1.00 Other 0.03 0.03 .40 1.00 0.00 0.03 .91 1.00 0.07 0.03 .02* .36 English -0.04 0.08 .60 1.00 0.01 0.08 .87 1.00 -0.01 0.08 .88 1.00 Italian 0.09 0.05 .06 1.00 0.05 0.04 .23 1.00 0.06 0.04 .21 1.00 Portuguese 0.18 0.04 <.001* <.001* 0.17 0.04 <.001* <.001* -0.01 0.04 .82 1.00 Serbian-Croatian 0.21 0.04 <.001* <.001* 0.17 0.03 <.001* <.001* 0.02 0.03 .65 1.00 Spanish 0.09 0.05 .06 1.00 0.05 0.04 .22 1.00 0.13 0.05 <.001* .06 Turkish 0.25 0.05 <.001* <.001* 0.27 0.05 <.001* <.001* 0.03 0.05 .55 1.00 Tamil 0.30 0.05 <.001* <.001* 0.12 0.05 .01* 0.22 -0.04 0.05 .35 1.00 Note. HE = higher education. *Significant at p<.05; NON-RESPONSE AND ATTRITION 33 Table 5: Multiple probit regression results predicting attrition by participant background and behaviour Predictor Outcome= P1-P4 attrition Outcome= Y1-Y4 attrition Outcome= Y5-Y7 attrition b SE p 6-FWER Holm corrected p b SE p 6-FWER Holm corrected p b SE p 6-FWER Holm corrected p Small Class (reference category= small class) - 0.02 0.04 .67 1.00 0.05 0.04 .23 1.00 0.02 0.03 .51 1.00 Gender (reference category= female) 0.00 0.02 .95 1.00 0.02 0.02 .51 1.00 - 0.04 0.02 .04* .65 Parental Higher Education - 0.01 0.02 .81 1.00 0.00 0.03 .96 1.00 - 0.07 0.02 <.001* .09 Neighbourhood Social Class 0.00 0.01 .70 1.00 0.00 0.01 .81 1.00 - 0.01 0.01 .57 1.00 Neighbourhood Familialism 0.00 0.01 .66 1.00 0.01 0.01 .48 1.00 0.00 0.01 .73 1.00 Language (reference category= German) Albanian 0.12 0.04 .01* .19 0.10 0.04 .02* 0.44 0.09 0.03 .01* .19 Other 0.13 0.03 <.001* <.001* 0.10 0.03 <.001* .03* 0.00 0.03 .95 1.00 English 0.02 0.07 .75 1.00 -0.10 0.08 .22 1.00 0.06 0.06 .32 1.00 Italian 0.00 0.04 .98 .98 -0.02 0.05 .70 1.00 0.04 0.04 .29 1.00 Portuguese 0.07 0.04 .10 1.00 0.01 0.05 .89 1.00 0.09 0.04 .02* 0.29 Serbian-Croatian 0.07 0.04 .07 1.00 0.11 0.04 <.001* .09 0.10 0.03 <.001* .04* Spanish 0.08 0.05 .09 1.00 0.08 0.05 .09 1.00 - 0.02 0.04 .70 1.00 Turkish 0.05 0.06 .33 1.00 0.04 0.06 .53 1.00 0.02 0.04 .71 1.00 NON-RESPONSE AND ATTRITION 34 Note. *Significant at p<.05. Tamil - 0.08 0.05 .14 1.00 -0.02 0.05 .72 1.00 - 0.06 0.04 .17 1.00 Parent reports Prosocial Behaviour 0.02 0.02 .33 1.00 0.00 0.02 0.88 1.00 0.00 0.02 0.84 1.00 Internalising - 0.03 0.02 .20 1.00 -0.03 0.03 .33 1.00 - 0.03 0.02 .22 1.00 ADHD - 0.01 0.02 .80 1.00 0.00 0.02 .87 1.00 0.00 0.02 .90 1.00 Non-aggressive Conduct Disorder - 0.07 0.04 .05* .91 -0.03 0.04 .46 1.00 - 0.02 0.04 .51 1.00 Aggression 0.00 0.03 .93 1.00 0.00 0.03 .95 1.00 0.00 0.03 .95 1.00 Self-reports Prosocial Behaviour - 0.02 0.06 .74 1.00 -0.06 0.06 .35 1.00 0.02 0.06 .73 1.00 Internalising 0.05 0.05 .32 1.00 0.05 0.05 .36 1.00 0.02 0.04 .68 1.00 ADHD - 0.04 0.07 .61 1.00 0.03 0.07 .68 1.00 - 0.03 0.06 .60 1.00 Non-aggressive Conduct Disorder 0.05 0.07 .46 1.00 0.05 0.07 .45 1.00 0.00 0.07 .96 1.00 Aggression - 0.05 0.08 .48 1.00 -0.05 0.08 .49 1.00 0.01 0.07 .90 1.00 Teacher reports Prosocial Behaviour - 0.01 0.01 .66 1.00 -0.01 0.01 .35 1.00 - 0.02 0.01 .09 1.00 Internalising - 0.02 0.01 .24 1.00 0.00 0.01 .74 1.00 0.00 0.01 .95 1.00 ADHD 0.02 0.01 .10 1.00 0.01 0.01 .71 1.00 0.00 0.01 .72 1.00 Non-aggressive Conduct Disorder 0.03 0.03 .38 1.00 0.04 0.03 .22 1.00 0.01 0.03 .68 1.00 Aggression 0.02 0.02 .41 1.00 0.03 0.02 .24 1.00 - 0.01 0.02 .53 1.00 NON-RESPONSE AND ATTRITION 35 Supplementary Table 1: Bivariate cross-correlation between behaviour problem measures 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. Parent reports 1.Prosocial Behaviour - 2.Internalising - 0.16 - 3.ADHD -0.2 0.42 - 4.Non-aggressive Conduct Disorder - 0.27 0.35 0.51 - 5.Aggression - 0.25 0.33 0.42 0.55 - Self-reports 6. Prosocial Behaviour 0.17 - 0.09 - 0.12 -0.06 -0.06 - 7.Internalising 0.09 8 0 -0.01 -0.02 0.22 - 8.ADHD - 0.05 0.09 0.1 0.1 0.1 -0.05 0.35 - 9. Non-aggressive Conduct Disorder - 0.06 0.07 0.14 0.14 0.12 -0.15 0.15 0.52 - 10.Aggression - 0.07 0.05 0.11 0.15 0.17 -0.08 0.36 0.57 0.53 - Teacher reports 11. Prosocial Behaviour 0.18 - 0.07 -0.1 -0.13 -0.06 0.15 0.06 -0.09 -0.15 -0.09 - 12.Internalising - 0.08 14 0.11 0.03 0 -0.06 0.04 0.08 0.09 0.07 -0.23 - 13. ADHD - 0.04 0.04 0.33 0.16 0.1 -0.11 0.05 0.18 0.2 0.15 -0.25 0.36 - 14. Non-aggressive Conduct Disorder - 0.02 0.04 0.24 0.18 0.12 -0.11 0.01 0.15 0.18 0.11 -0.32 0.38 0.63 - 15. Aggression - 0.05 0.03 0.21 0.13 0.17 -0.06 0.01 0.16 0.15 0.12 -0.3 0.34 0.56 0.69 - NON-RESPONSE AND ATTRITION 36 Note. Cross-informant correlations on same phenotype. NON-RESPONSE AND ATTRITION 37