Riassunto esame Digital media , Prof. Ceravolo Flavio, libro consigliato Digital Social Research, Veltri Giuseppe

Online Experiments

The main strategic advantage of an online experiment is the increasing possibility for generalization, greater statistical power, and possibly quality of the data produced. Web studies, by having larger samples, usually have greater power than lab studies. The main drawback of an online experiment compared to a laboratory one is the lack of full environmental control. There are a few cons, like MULTIPLE SUBMISSIONS: they can be avoided or controlled by collecting personal identification items.

The main trade-off between online and lab experiment is that of exchanging greater generalizability and data power for less experimental control > in fact, experiments are often repeated with the same outcome measures both online and in lab.

Other advantages of online experiments are:

• The speed of data collection
• High degree of automation of the experiment (low maintenance)
• A wider sample both in size and wider geographical variations in participants.

There are other potential

limitations that are apparent depending on the research topic. These include the following factors: - web experiments' dependency on computers and networks having psychological, technical and methodological implications; - from a cognitive POV, participant at computers will likely be subject to self-actualization and other influences in computer-mediated communication; - the temporal duration of online experiments is shorter compared to laboratory ones > the most common duration being around twenty minutes. Online experiments exert a higher cognitive toll on respondents than laboratory settings. The reason might be that it is easier to exercise focus and attention in an environment that is designed to allow it - the laboratory - than in other environments, such as the home. * Online experiments as field experiments: online experiments can also take the form of field experiments. They are effective in collecting evidence on actual people's behavior, because people are directly

observed in real context and participants are usually unaware that they are part of a social experiment until the end > this retain the ability to understand the effect of a target variable that is “manipulated”, and also ethical problems can be arised. At the same time, experiments that are about native online behavior and social phenomena have appeared. These are field experiments where the field is digital. Questions about the ethical viability of these experiments are being raised, especially because private research companies use them.

QUANTITATIVE DATA ANALYSIS RELOADED

In digital social scientific research, there are three types of quantitative analysis being applied and developed above others:

1. Statistical methods of data dimensionality reduction
2. Quantitative methods of analyzing relational data, namely network analysis
3. Quantitative methods of analyzing texts as unstructured data.

In this context, it is useful to distinguish between DEPENDENCE and INTERDEPENDENCE methods,

In analisi multivariata, esistono diversi metodi. In un metodo dipendente, una variabile o un insieme di variabili viene identificato come variabile dipendente da prevedere o spiegare da altre variabili indipendenti (analisi di regressione multipla, analisi discriminante e analisi congiunta). Questo metodo si basa sulle teorie di causalità regolare. Un metodo di interdipendenza è un metodo in cui nessuna singola variabile o gruppo di variabili viene definito come indipendente o dipendente (analisi di cluster, analisi dei fattori, analisi dei componenti principali e scaling multidimensionale). > ANALISI DEI FATTORI: il suo scopo principale è definire la struttura sottostante della variazione esistente tra le variabili nell'analisi. Ha due approcci: prima, i dati possono essere analizzati senza idee preconcette sulle costruzioni sottostanti (variabili latenti) = analisi fattoriale esplorativa. Secondo, quando si ha una comprensione delle costruzioni sottostanti i dati, possono essere analizzati utilizzando l'analisi fattoriale confermativa (CFA): un test o un'ipotesi specifica.

About the structure or the number of dimensions underlying a set of variables is performed. The outcome of the factor analysis is twofold:

1. DATA SUMMARIZATION: describe the data in a much smaller number of concepts than the original individual variables
2. DATA REDUCTION: derive an empirical value (factor score) for each dimension (factor) and then substituting this new value for the original values of the processed variables.

FA investigates whether a number of variables of interest are related through some linear function to a smaller number of unobservable factors (latent variables or constructs) > is able to summarize data in more manageable formats.

CLUSTER ANALYSIS: whereas factor analysis reduces the number of variable by grouping them into a smaller set of factors, cluster analysis reduces the number of observations or cases by grouping them into a small set of clusters > the resulting clusters should exhibit high internal (within-cluster) homogeneity and high external (between-cluster) heterogeneity.

Cluster methods can be split into hierarchical and non-hierarchical methods. The interdependence methods tend to be more frequent in digital data: we are moving from a case in which we need to ensure generalizable and solid inferences from small datasets to a case in which we have the problem of selecting and reducing the amount of information that we have from large amounts of data. *Conventional and Computational Approaches* The role of digital data (especially big data) has had an impact on social science research methods as a result of the computational and algorithmic turn. We can contrast the difference between the "two cultures of modeling": - DATA-MODELING CULTURE: is about evaluating the value of parameters from the data, and after that the model is used for either information or prediction. - ALGORITHMIC MODELING: there is a shift from data models to the properties of algorithms, in order to predict responses. The majority of quantitative analytical methods used in social science research can be categorized into either conventional or computational approaches.

Science belongs to the data-modeling culture, but as the complexity of datasets increases, there could be a multiplicity of models that are still a good fit for the dataset, even though they might express different relationships between variables. We often apply models that are global, without considering the possibility of local models for a given segment of the dataset.

Social science methods give emphasis on variable selection, but they have a problem of model selection, and they aim at hypothesis testing while computer science methods focus on prediction. The goals of the social sciences approach are different from those of computer science.

Computational Social Science (CSS): The interdisciplinary investigation of the social universe on many scales, ranging from individual actors to the largest groupings, through the medium of computational techniques in contrast to traditional statistical methods.

Model-based recursive partitioning: This approach is an improvement on the use of CART (methods based on a

The purely data-driven paradigm. Without making use of a predefined statistical model, such algorithmic methods search recursively for groups of observations with similar values to the response variable by building a tree structure. They are purely descriptive) > This approach works through following steps:

1. A parametric model is defined to express a theory-driven set of hypotheses
2. This model is evaluated by the model-based recursive partitioning algorithm, which checks whether other important covariates have been omitted that would alter the parameters of the initial model
3. The model-based recursive partitioning algorithm finds different patterns of associations between the response variable and other covariates that have been pre-specified in the parametric model. In other words, it creates different versions of the model in terms of beta estimation, depending on different important values of covariates.

The complexity, quantity and availability of digital data have highlighted the need

social media interactions, email exchanges, etc. Network analysis allows us to understand thestructure and dynamics of social relationships, identifying key actors, communities, andinfluencers.DataData is the fuel of computational social science. The availability of large-scale datasets hasrevolutionized the way social scientists can study human behavior and social phenomena. Bigdata, combined with advanced computational techniques, allows for the analysis of massiveamounts of information, revealing patterns and trends that were previously inaccessible. Thedata can come from various sources, such as social media platforms, government databases,sensor networks, and online surveys. The challenge lies in effectively managing, cleaning, andanalyzing these vast datasets, as well as ensuring their ethical use.ModelingModeling is a fundamental aspect of computational social science. It involves creatingmathematical or computational representations of social phenomena and processes. Thesemodels can range from simple mathematical equations to complex agent-based simulations.Models allow researchers to test hypotheses, explore different scenarios, and make predictionsabout social behavior. They can also help in understanding the underlying mechanisms thatdrive social phenomena. However, modeling in computational social science is not withoutits challenges. It requires careful consideration of assumptions, parameterization, andvalidation against real-world data.ConclusionComputational social science offers exciting opportunities for social scientists to study socialphenomena in new and innovative ways. By leveraging computational techniques, such asnetwork analysis, data analysis, and modeling, researchers can gain new insights and makeimportant contributions to the field. However, it is important to recognize the methodologicalchallenges and ethical considerations that come with this approach. With careful andrigorous research practices, computational social science can bring about transformativechanges in our understanding of society.

In the attempt at understanding a hyperconnected world. In quantitative social research methods the fundamental assumption is the independence of observations on individual units (methodological individualism); network analysis, instead, adopts a constructionist/relationist approach - the interest is in the relationships between subjects, because units are not acting independently from one another; instead, they influence each other.

The most basic feature of network measurement is the use of structural or relational information to study test theories. All these concepts are quantified by considering the relations measured among the actors in a network.

Social network data require measurements on ties among social units (or actors). The network perspective takes into consideration that networks can have direct and indirect effects on actor's behavior or outcomes and that actors can belong to several overlapping social networks. Because network measurements give rise to data that are

unlike othersocial and behavioral science data, an entire body of methods has been developed, withanalytical procedures to determine how the system of relationships between actors behaves,and statistical methods to test the appropriateness of the propositions about such a system.Social scientists study social networks using social network analysis (SNA).

Key Concepts

We can start defining a network as a set of entities (vertices or nodes) and a set of edgesthat indicate connections for relations between the vertices. In SNA, we talk of actors(nodes) and ties (edges). We call a sociogram a visual representation of a social network.Two main elements of a social network are:

1. ACTORS: they can be of different types (people, organizations, communities) and canhave attributes, and we need to distinguish between non-graph-theoretic attributes (intrinsicto the actor, like age, gender) and graph-theoretic attributes, which are actor attributesgenerated from their pos