Traiter ses données proprement : vers un meilleur usage du data cleaning

Data cleaning has long been shown to be an essential step in data analysis but its application is not systematic and varies between researchers. The aim of the present article is to study the effect of the different elements of data cleaning and to give recommendations. The steps considered are normality, outliers and missing values. Firstly, we advise a close examination of the data and its normality with a view to transformation. Secondly, we look at outliers and their treatment (trimming and winsorizing). Lastly, we turn our attention to missing values and how to deal with them depending on their randomness. We also cover the importance of visualizing data before analysis. Finally, we illustrate the effect of these data cleaning practices on a real data set. We show that data cleaning can lead to a significant result becoming non-significant and therefore demonstrate the importance of these steps before starting data analysis.     

Exploratory Approaches for Studying Social Interactions,Dynamics, and Multivariate Processes in Psychological Science

In this article, I argue for the need of more use of exploratory techniques to identify dynamics in social interactions. I describe several approaches as they are applied to multivariate time series data. The first approach is an algorithm that searches for periods of variability and stability at the individual level as well as for patterns of overlap in such periods between the two individuals in a couple. These patterns describe the daily ups and downs in the couples' affect and are predictive of the state of the couples 1 to 2 years later. The second approach, hierarchical segmentation, is based on the idea of partitioning the time series in segments with distinct data patterns. In the case of data from dyads, as in the illustration, the patterns can be compared in terms of coherence between the 2 individuals in the dyad. The third approach is based on network analysis, and its use is shown as a method to examine data transitions at the individual and dyadic level as well as system-wide coherence in multivariate systems. For each approach, I provide examples of its use with empirical data. The article ends with general guidelines and recommendations for researchers interested in using exploratory methods as a way to examine psychological processes.     

网民对“人祸”事件的道德情绪特点——基于微博大数据研究

本研究采用大数据研究方法,对爬取的"动车事故"发生后40天内的94,562条相关微博进行情感分析,以探讨网民对"人祸"的道德情绪特点,同时对不同群体情绪表达差异进行探讨。结果发现:(1)网民对于动车事故主要表达的道德情绪有:愤怒、鄙视、厌恶、同情和爱。(2)包含不同道德基础的事件与不同的道德情绪相关联;(3)对于愤怒、厌恶和鄙视,男性普遍有更高的表达倾向和表达强度,而女性更倾向于表达爱和同情且强度更高;(4)对于爱和同情,团体VIP用户组表达的可能性和强度都高于其他用户;个体VIP用户比非VIP用户更可能表达愤怒、鄙视和厌恶,而团体VIP用户表达这类情绪的强度最小。研究表明,虚拟网络中人们道德情绪特点依然符合道德基础理论;不同群体在表达道德情绪时的差异性是对道德基础理论相关研究的补充。总言之,数据挖掘技术和情感分析方法是进行情绪研究的有效手段。     

The Entangling of Problems,Solutions and Markets: On Building a Market for Privacy

An increasing number of public problems have been subject to market-based interventions as solutions. However, the relationship between problems and solutions in market-based interventions is complex. On occasions solutions are reformulated as understanding of the nature of the problem are advanced. Alternatively, problems are reconfigured to fit a standard solution. Or solutions are said to generate numerous new problems. The complex entangling of problems, solutions and markets can be explored by focusing on the field of online privacy. The complexities of this field can be analysed through three science and technology studies (STS) treatments of problems and solutions. First, issue-problematisation can be used to understand the ways in which a problem is assembled and comes to form the focus for discussion as an issue. Second, a paradigm-exemplar approach can assess the extent to which a particular coupling of problem and solution becomes an exemplar for others to draw on. Third, ontological constitution provides a focal point for analysing the ways in which the very nature of entities involved in problems, solutions and markets can be considered. There is analytic utility in each of these approaches in engaging with online privacy regulation and the emerging role of start-ups in introducing new privacy products to an emerging market place. However, these STS approaches leave us with questions regarding the relationship between market solutions and publics, the co-ordination of entities in market solutions and the normative questions that arise from entangling markets, problems and solutions.     

Data Labours: How the Sequence Databases GenBank and EMBL-Bank Make Data

What actually happens inside genetic databases, how do they work upon data and who does this work? While they have become central tools for doing science, not much is known about the work that goes on inside these vital infrastructures. Ethnographic explorations of two of the world’s largest nucleotide sequence databases, GenBank and the European Molecular Biology Laboratory’s EMBL-Bank, reveal manifold goings-on. Like most infrastructural work, it is modest and invisible routines that build and maintain the vast interconnected suite of bioinformational resources. Data curators construct organisms out of sulphuric sludge, dataflow engineers as self-styled “genetic information plumbers” keep the data deluge flowing, and a data submissions support assistant manages to make room for care amidst this deluge. Taken together, these data labours render tangible the modest and processual aspects of data infrastructure while also revealing the databases to be situated and lively spaces of convergence. Inventively analysing data labours paves surprising ways for encountering and making sense of databases, data and the work they do. Here, practices of natural history, like specimen-making and curation, are continued by other means while the assembly of sludge sheds light on the absences and deletions which mystify infrastructural maintenance work.     

Rising High or Falling Deep? Pathways of Self‐Esteem in a Representative German Sample

In many longitudinal studies, self‐esteem has been shown to increase up until around age 50 or 60 and to decrease thereafter. These studies have also found substantial inter‐individual differences in the intra‐individual development of self‐esteem. In the current study, we examined whether this variation in change could be attributed to underlying latent classes of individuals following different trajectories of self‐esteem development over time. By applying general growth mixture modelling to data from the representative German pairfam study (N = 12 377), four latent classes of self‐esteem development across five years were extracted. Based on their mean levels, trajectories, and variability, individuals in the latent classes could be described as having (a) constant and stable high self‐esteem (29.00% of the sample), (b) constant but variable moderate self‐esteem (31.69%), (c) increasing and stabilizing self‐esteem (15.13%), and (d) decreasing and variable self‐esteem (24.18%). Furthermore, these latent classes differed in accordance with findings of prior research on self‐rated, partner‐rated, and objective correlates of the domains of health and well‐being, partner relationships, and occupational status. Thus, the current study shows that inter‐individual variation in intra‐individual change in self‐esteem is not random but reflects specific individual trajectories, or pathways, of self‐esteem. Copyright © 2016 European Association of Personality Psychology     

Introducing human factors in pedestrian crossing behaviour models

The objective of this research is the development of pedestrian crossing choice models on the basis of road, traffic and human factors. For that purpose, a field survey was carried out, in which a panel of 75 pedestrians were asked to take 8 short walking trips (each one corresponding to a different walking and crossing scenario) in the Athens city centre in Greece, allowing to record their crossing behaviour in different road and traffic conditions. The same individuals were asked to fill in a questionnaire on their travel motivations, their mobility characteristics, their risk perceptions and preferences with respect to walking and road crossing, their opinion on drivers, etc. The walking and crossing scenarios’ data were used to develop mixed sequential logit models of pedestrian behaviour on the basis of road and traffic characteristics. The modelling results showed that pedestrian crossing choices are significantly affected by road type, traffic flow and traffic control. The questionnaire data were used to estimate human factors (components) of pedestrian crossing behaviour by means of principal component analysis. The results showed that three components of pedestrian crossing behaviour emerge, namely a “risk-taking and optimisation” component reflecting the tendency to cross at mid-block in order to save time, etc., a “conservative” component, concerning individuals with increased perceived risk of mid-block crossing, who also appear to be frequent public transport users, and a “pedestrian for pleasure” component, bringing together frequent pedestrians, walking for health or pleasure, etc. The introduction of these components as explanatory variables into the choice models resulted in improvement of the modelling results, indicating that human factors have additional explanatory power over road and traffic factors of pedestrian behaviour. Therefore, the development of integrated choice and latent variables models appears to be an appropriate field for further research.     

Modeling cyclist acceleration process for bicycle traffic simulation using naturalistic data

Cycling is a healthy and sustainable form of transportation. The recent increase of daily cyclists in Sweden has triggered broad interest in finding how policies and measures may facilitate the planning of bicycle traffic in the urban area. However, in comparison to car traffic, bicycle traffic is still far from well understood. This study is part of the research effort to investigate microscopic cyclist behavior, model bicycle traffic and finally build a simulation tool for applications in transport planning. In particular, the paper focuses on representing bicycle movements when the cyclist doesn’t interact with others. The cyclist acceleration behavior is modeled using naturalistic GPS data collected by eleven recruited commuter cyclists. After filtering the large amount of data, cyclist trajectories are obtained and acceleration profiles are abstracted. A mathematical model is proposed based on the dataset, and three model forms are estimated using the maximum likelihood method with Laplace and Normal error terms. While the model with more parameters shows superior performance, the simplified ones are still capable of capturing the trends in the acceleration profiles. On the other hand, the study also introduces social economic characteristics of cyclists to explain the model parameters and they show significant effects. However, the cyclist population being investigated in the study is still limited, and more convincing results can be obtained when the data collection effort is extended to larger population with more variable cyclist characteristics.     

Field study on the behavior of right-turning vehicles in Malaysia and their contribution on the safety of unsignalized intersections

Behavior of right turning vehicles in the context of safety is characterized by their use of turning indicators and compliance with the stop rule. They are influence by the width of the carriageway and the variation in volume on the major road with respect to the traffic moving in the near and far side direction. Other factors affecting the behavior are the speed and spacing between vehicles moving on the major road. Lack of adequate past knowledge on the effect of geometric variation in terms of road width and directional variation in volume on the safety of unsignalized intersections have provided the motivation for this study. This paper focuses on the many factors that affect the behavior of right-turning vehicles resulting into conflicts. A brief account of the unique indigenous maneuver termed as the “Weaving Merging Right Turn” (WMRT) is provided and its effectiveness with respect to conventional right turn is evaluated. Data of 39,016 vehicles collected on 10 sites between January and June 2014 was analyzed. Multiple accidents were observed only on sites which had near side traffic volume greater than far side traffic volume. This result remains consistent with sites having single as well as multiple lanes per direction on the major roads. The number of conflicts for vehicles performing the WMRT was 2.5 times less as compared to the conventional right turn. Moreover WMRT was found to be the maneuver of choice for right turning motorcyclists with 60% of them opting for it over the conventional right turn on intersections having major road width less than 9 m. None of the motorcyclists, which were involved in a traffic conflict, were observed to use their turning indicator. Moreover none of the motorcyclists, which experienced a traffic conflict, were found to comply with the stopping rule at sites with major road width less than 9 m. On sites with major road width greater than 9 m, 45% of motorcyclists, involved in a traffic conflict, complied with the stopping rule as compared to 79% by vehicles other than motorcycles.     

Using naturalistic data to assess e-cyclist behavior

In Europe, the use of electric bicycles is rapidly increasing. This trend raises important safety concerns: Is their use compatible with existing infrastructure and regulations? Do they present novel safety issues? How do they impact other traffic? This study sought to address these concerns, using instrumented electric bicycles to monitor e-cyclists’ behavior in a naturalistic fashion. Data was collected from 12 bicyclists, each of whom rode an instrumented bicycle for two weeks. In total, 1500 km worth of data were collected, including 88 critical events (crashes and near-crashes). Analysis of these critical events identified pedestrians, light vehicles and other bicycles as main threats to a safe ride. Other factors also contributed to crash causation, such as being in proximity to a crossing or encountering a vehicle parked in the bicycle lane. A comparison between electric and traditional bicycles was enabled by the availability of data from a previous study a year earlier, which collected naturalistic cycling data from traditional bicycles using the same instrumentation as in this study. Electric bicycles were found to be ridden faster, on average, than traditional bicycles, in addition to interacting differently with other road users. The results presented in this study also suggest that countermeasures to bicycle crashes should be different for electric and traditional bicycles. Finally, increasing electric bicycle conspicuity appears to be the easiest, most obvious way to increase their safety.