Bayesian networks for imputation

Summary.  Bayesian networks are particularly useful for dealing with high dimensional statistical problems. They allow a reduction in the complexity of the phenomenon under study by representing joint relationships between a set of variables through conditional relationships between subsets of these variables. Following Thibaudeau and Winkler we use Bayesian networks for imputing missing values. This method is introduced to deal with the problem of the consistency of imputed values: preservation of statistical relationships between variables ( statistical consistency ) and preservation of logical constraints in data ( logical consistency ). We perform some experiments on a subset of anonymous individual records from the 1991 UK population census.     

Multilevel modelling of refusal and non-contact in household surveys: evidence from six UK Government surveys

Summary.  We analyse household unit non-response in six major UK Government surveys by using a multilevel multinomial modelling approach. The models are guided by current conceptual frameworks and theories of survey participation. One key feature of the analysis is the investigation of the extent to which effects of household characteristics are survey specific. The analysis is based on the 2001 UK Census Link Study, which is a unique source of data containing an unusually rich set of auxiliary variables. The study contains the response outcome of six surveys, linked to census data and interviewer observations for both respondents and non-respondents.     

Obstacles to involving children and young people in foster care research

ABSTRACT Despite increasing recognition of the importance of including the perspectives of children and young people in care in alternative care research, in practice this is not always a straightforward matter. This paper describes the recruitment of disruptive young people in care under the jurisdiction of the South Australian statutory authority to three studies on placement instability. Non‐response rates of 72.5% and 82% are reported. A large number of subjects were excluded because agency social workers did not cooperate with the project, and more subjects were excluded for reasons which suggest high levels of distress in this population. The dilemma of providing a voice to distressed subjects when distressed subjects are excluded from research is discussed, and the appointment of an independent representative for children in care to review research proposals and to negotiate research access to children is proposed.     

Interval estimation for log-linear models with one variable subject to non-ignorable non-response

Summary.  Log-linear models for multiway contingency tables where one variable is subject to non-ignorable non-response will often yield boundary solutions, with the probability of non-respondents being classified in some cells of the table estimated as 0. The paper considers the effect of this non-standard behaviour on two methods of interval estimation based on the distribution of the maximum likelihood estimator. The first method relies on the estimator being approximately normally distributed with variance equal to the inverse of the information matrix. It is shown that the information matrix is singular for boundary solutions, but intervals can be calculated after a simple transformation. For the second method, based on the bootstrap, asymptotic results suggest that the coverage properties may be poor for boundary solutions. Both methods are compared with profile likelihood intervals in a simulation study based on data from the British General Election Panel Study. The results of this study indicate that all three methods perform poorly for a parameter of the non-response model, whereas they all perform well for a parameter of the margin model, irrespective of whether or not there is a boundary solution.     

Estimation of mean vector in presence of non-response

Simultaneous estimation of means of several variables is considered for finite population in presence of non-response. Two types of nonresponses (partial and complete) are considered using the technique of sampling and subsampling with equal probabilities without replacement. The optimum sample size and the optimum value of subsampling fraction to be repeated from the nonresponding units of the sample have been obtained for fixed survey budget.     

Non-response in social networks: The impact of different non-response treatments on the stability of blockmodels

Discerning the essential structure of social networks is a major task. Yet, social network data usually contain different types of errors, including missing data that can wreak havoc during data analyses. Blockmodeling is one technique for delineating network structure. While we know little about its vulnerability to missing data problems, it is reasonable to expect that it is vulnerable given its positional nature. We focus on actor non-response and treatments for this. We examine their impacts on blockmodeling results using simulated and real networks. A set of ‘known’ networks are used, errors due to actor non-response are introduced and are then treated in different ways. Blockmodels are fitted to these treated networks and compared to those for the known networks. The outcome indicators are the correspondence of both position memberships and identified blockmodel structures. Both the amount and type of non-response, and considered treatments, have an impact on delineated blockmodel structures.     

Item non-response in open-ended questions: Who does not answer on the meaning of left and right?

The left-right ideological dimension plays a central role in political science research. The meaning of left and right is often based on analyses of responses to open-ended survey questions. In the 2008 German General Social Survey, item non-response to two particular open-ended questions on the meaning of left and right was considerable and calls for analyses of the value of the data have been received. In our paper we investigate item non-response to these two open-ended questions in terms of cognitive abilities, motivational aspects, and finally, in relation to the left-right self-placement scale itself. Item non-response in the open-ended questions is not based on randomness alone, and analyses show some significant systematic results. Thus, the open-ended questions seem to be biased and the validity of this data is questionable in some respect. In this context, left-right self-placement of non-respondents and “cognitive don’t knowers” need to be given careful consideration.     

Actor non-response in valued social networks: The impact of different non-response treatments on the stability of blockmodels

Social network data usually contain different types of errors. One of them is missing data due to actor non-response. This can seriously jeopardize the results of analyses if not appropriately treated. The impact of missing data may be more severe in valued networks where not only the presence of a tie is recorded, but also its magnitude or strength. Blockmodeling is a technique for delineating network structure. We focus on an indirect approach suitable for valued networks. Little is known about the sensitivity of valued networks to different types of measurement errors. As it is reasonable to expect that blockmodeling, with its positional outcomes, could be vulnerable to the presence of non-respondents, such errors require treatment. We examine the impacts of seven actor non-response treatments on the positions obtained when indirect blockmodeling is used. The start point for our simulation are networks whose structure is known. Three structures were considered: cohesive subgroups, core-periphery, and hierarchy. The results show that the number of non-respondents, the type of underlying blockmodel structure, and the employed treatment all have an impact on the determined partitions of actors in complex ways. Recommendations for best practices are provided.     

The effects of income imputation on microanalyses: evidence from the European Community Household Panel

Summary.  Social surveys are usually affected by item and unit non-response. Since it is unlikely that a sample of respondents is a random sample, social scientists should take the missing data problem into account in their empirical analyses. Typically, survey methodologists try to simplify the work of data users by 'completing' the data, filling the missing variables through imputation. The aim of the paper is to give data users some guidelines on how to assess the effects of imputation on their microlevel analyses. We focus attention on the potential bias that is caused by imputation in the analysis of income variables, using the European Community Household Panel as an illustration.