Study Of Simulation And Noise Robustness For Non-randomized Response Model Towards The Sensitive Issue

In social survey, we usually need survey some sensitive issues that are closely related to personal privacy and benefit. If these sensitive issues are asked directly, the respondents may tend to give the wrong answers and even refuse to reply. Non-randomized response technique can effectively protect the privacy of respondents and encourage respondents to make real answers. The greatest advantage of the technique is repetitive. Although the technique has made a major breakthrough, some respondents do not understand the rules of the survey or are inclined to make errors in reality. Therefore these situations will make survey results failure. These situations of interferential survey results have already existed but due to the complexity of the actual investigation it cannot be studied well. This paper adopts a new way to study the non-randomized response model of the questionnaires to address the preceding interferential situation.The paper mainly studies non-randomized response model which tolerates the error in the actual questionnaires(i.e. the study of its noise robustness). It could not be an effective noise robustness study if using a general investigation method to explore the noise robustness of non-randomized response model,even though it has enough time and money. Therefore, the article which comes from a new perspective applies computational simulation to study the noise robustness of non-randomized response model.Subsequently, through computational simulation experiments, the model successful-ly realizes the process of questionnaire survey on the same sensitive problem. This paper simulates the results of non-randomized response models which under the circumstances of success and failure. Computational simulation experiment is proved to be an effective and reliable method, and it can be applied to study noise robustness of non-randomized response model in the case of the questionnaire error.More importantly, according to the specific circumstances of the questionnaire error, the paper designs the corresponding non-randomized response model. This paper also deduces the theoretical mean and variance of sensitive issues for cross model, triangular model and parallel model under the circumstances of errors. Based on effective simulation experiments, we implement computational simulation to achieve questionnaire for these four non-randomized response models in the case of error. At the same time, experiments find the noise robustness intervals for every model. The noise robustness interval is an important indicator for noise robustness of model. The accuracy of these non-randomized response models is enough to be proved, according to the simulation results. Next, the article discusses the influence of model parameters on noise robustness. The paper shows that noise robustness of cross model and variant parallel model is better than triangle model and parallel model, but the stability of variant parallel model and triangle model becomes worse in comparison with the others. For each model, the varying parameters seriously affect its robustness against noise. This work provides a theoretical reference and guiding principles for non-randomized response model in reality.