ARTICLE

https://doi.org/10.1057/s41599-024-03037-x OPEN

Response Item Network (ResIN): A network-based

approach to explore attitude systems
Dino Carpentras1 ✉, Adrian Lueders2 & Michael Quayle 3

Belief network analysis (BNA) refers to a class of methods designed to detect and outline
structural organizations of complex attitude systems. BNA can be used to analyze attitude-
1234567890():,;

structures of abstract concepts such as ideologies, worldviews, and norm systems that inform
how people perceive and navigate the world. The present manuscript presents a formal
specification of the Response-Item Network (or ResIN), a new methodological approach that
advances BNA in at least two important ways. First, ResIN allows for the detection of attitude
asymmetries between different groups, improving the applicability and validity of BNA in
research contexts that focus on intergroup differences and/or relationships. Second, ResIN’s
networks include a spatial component that is directly connected to item response theory
(IRT). This allows for access to latent space information in which each attitude (i.e. each
response option across items in a survey) is positioned in relation to the core dimension(s) of
group structure, revealing non-linearities and allowing for a more contextual and holistic
interpretation of the attitudes network. To validate the effectiveness of ResIN, we develop a
mathematical model and apply ResIN to both simulated and real data. Furthermore, we
compare these results to existing methods of BNA and IRT. When used to analyze partisan
belief-networks in the US-American political context, ResIN was able to reliably distinguish
Democrat and Republican attitudes, even in highly asymmetrical attitude systems. These
results demonstrate the utility of ResIN as a powerful tool for the analysis of complex attitude
systems and contribute to the advancement of BNA.

1 Computational Social Science, ETH Zürich, Zürich, Switzerland. 2 Department of Communication Science, University of Hohenheim, Stuttgart, Germany.
3 Department of Psychology, University of Limerick, Limerick, Ireland. ✉email: [email protected]

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 1

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Studying belief systems link’s weight. This weight is usually calculated as a correlation
A reinvigorated interest in attitudes and the need for adequate (Boutyline and Vaisey, 2017), although many alternatives are
methodologies. Researching attitudes is a central goal of the social possible, such as the Kullback-Leibler distance (Kertzer et al.
sciences as it can inform a wide range of pressing social phenomena 2019).
(Lüders et al. 2023). Not only do attitudes reflect internal repre- Belief networks that are obtained in such ways can be visually
sentations of the outer world; but they also permit individuals to inspected based on standard visualization techniques from
strategically interact with their social environment as they often network analysis (Krempel, 2011). This initial step offers a quick
convey information about latent concepts such as worldviews and and intuitive (but imprecise) way to explore the structures of
identities (Quayle, 2020; Klein et al. 2007). For instance, when belief networks. Visual inspection can be used to identify clusters
navigating social contexts that are less explicit or clearly defined, (DellaPosta, 2020), peripheral nodes, and positive or negative
such as social media, individuals rely on meaningful cues to evaluate relationships between items (Boutyline and Vaisey, 2017; Brandt
and categorize their social environment (Postmes et al. 2005), and et al. 2019), or used to (roughly) estimate the number of factors in
there is good reason to believe that attitude expressions in the form the dataset and their relationship to each other (Kan et al. 2020).
of tweets, likes, hashtags, user-bios, and other available affordances A more precise analysis of belief networks can be realized based
serve exactly such functions (Lüders et al. 2022; McGarty et al. on quantitative techniques developed in network analysis (Serrat,
2014). Hence, by engaging with other users through attitude 2017). For example, it is possible to estimate the size of a cluster
expressions, individuals strategically locate themselves within their (i.e. how many nodes compose it), how separate it is from the rest
wider social environment. Conversely, by using attitudes as social of the network, which are the most central nodes, etc. (Boutyline
cues, individuals can infer information and make judgements about and Vaisey, 2017; DellaPosta, 2020; Serrat, 2017). Indeed, it is
others (Lüders et al. 2022). common practice to start with a visual and qualitative inspection
Network-based methods provide novel opportunities to study of a network and then analyze its properties more deeply based
the structural basis of complex attitude systems, therefore helping on quantitative techniques.
researchers reassessing their meaning for abstract concepts like Another major advantage of BNA is that despite being a
worldviews, identities, and ideologies. Furthermore, by operatio- network-based method, BNA is computationally fast, hence
nalizing attitudes at the system level, attitude networks can foster allowing the analysis of large datasets. Indeed, in many network-
the understanding of social influence and attitude change based methods the time required for the calculation grows non-
(Carpentras et al. 2022). These advantages hold theoretical as linearly with the number of nodes (Wakita and Tsurumi, 2007).
well as practical value. For instance, the research found that This means that several techniques that model people as nodes
shared attitude (dis)agreement lies at the core of group cannot be applied on big datasets, as they will require excessive
constructions among climate change advocates and sceptics computational time. Conversely, modelling items as nodes allows
(Bliuc et al. 2015) and that newly emerging belief networks in the for a more parsimonious network structure and hence more
wake of the Covid-19 pandemic successfully predicted health computational resources.
measure compliance (Maher et al. 2020).
Limitations of belief network analysis. While BNA has been
Belief Network Analysis for studying attitude systems. Belief proven to be a very powerful and promising tool, like any other
network analysis (BNA) is a relatively young approach for method, it also suffers from various limitations. The first stems
exploring concepts such as attitudes, goals, or values as well as from the techniques that are used for calculating the weights of
their mutual interplay. In this paper, we loosely use the term to links that connect two nodes. As previously mentioned, link
refer to the family of approaches that treat beliefs (or attitudes, weights are typically calculated based on linear correlation
opinions, etc.) as nodes and build links between them using linear (Boutyline and Vaisey, 2017). However, these methods can only
correlation (e.g. Converse, 1964; Boutyline and Vaisey, 2017). reflect monotonic (i.e. only increasing or only decreasing) rela-
Belief networks thus differ from network methods that represent tionships (De Winter et al. 2016). Let us consider a practical
people as nodes (e.g. retweet networks, follower networks). While example in which survey participants from both extremes of the
social network methods usually try to provide a structure of the political spectrum (i.e. left, right) score high on a fictitious “item
social connections among the studied population, belief networks 1.” Let us also suppose that moderates (i.e. people falling between
provide information about the structural organization of belief the two poles) score lower on that same item. The data resulting
systems (Boutyline and Vaisey, 2017; DellaPosta, 2020; Brandt from this example would follow a U-shaped curve (Fig. 1a).
et al. 2019; Kertzer et al. 2019). Since BNA are network-based Despite the fact that this relation is characterized by very low
methods, they allow for quick visual inspections of belief systems, random fluctuations (i.e. based on one variable one can precisely
as well as precise and quantitative analysis of their underlying predict the other), the calculated correlation coefficient would be
attitudinal architecture. zero. Thus, by looking only at the correlation value, most people
Even though some BNA techniques can be applied to diverse will erroneously conclude that no relationship exists between item
types of data (e.g. unstructured data from social media), they are 1 and the left-right spectrum.
usually used to process survey data, modelling items as nodes The simulated relationships depicted in Fig. 1b–d reveal a
while edges are derived from some form of inter-item correlation different problem as very different relationships produce the same
(Boutyline and Vaisey, 2017; DellaPosta, 2020; Brandt et al. 2019; correlation value (r = 0.45). Figure 1b, c show the case in which
Kertzer et al. 2019). The process for producing a network is all people score the same value except for people on the extreme
typically simpler than the steps required for many other left (1b) or the extreme right (1c). Therefore, these two items
techniques extracting structural features of multidimensional characterize very precisely those people who are falling at the
relationships between items. For example, common techniques extremes of the spectrum. Instead, Fig. 1d shows a more classical
such as factor analysis aim to summarize a larger set of items into linear relationship between variables but with much higher
a smaller set of factors. In contrast, BNA outlines a “map” of uncertainty (i.e. we cannot precisely infer one variable’s value
relationships between items, however, without changing their from the other, due to random fluctuations). Unfortunately, just
qualitative interpretation1. To achieve this, BNA methods simply by looking at the link between two nodes (i.e. the information
calculate the strength of between-item relationships for each pair provided by BNA), we will not be able to understand which of
of items and incorporate this information in the network as each these three patterns is producing the calculated correlation value.

2 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

Fig. 1 Simulations of possible different response patterns. a Despite having no random variations in the relationship, the correlation between the two
variables is zero. b–d Three very different relationships producing the same correlation (r = 0.45).

Some alternative methods have been suggested to address such negative towards items B and C. We would further conclude that
problems. For example, instead of using correlation to estimate item D has no connection to political self-identification (i.e. item D
link weights, some authors used the Kullback-Leiber distance has negligible correlations with the other items).
(Kertzer et al. 2019). However, this does not solve the problem As the reader may notice, this interpretation implicitly assumes
that different curves will result in the same coefficient (thus, being that the attitudes held by people on the extreme left and right are
indistinguishable during the network analysis phase). Indeed, this symmetric: the more positive left-wingers are towards an item
ambiguity does not stem from the use of a specific correlation or (e.g. a certain policy) the more negative are right-wingers, and
distance, but from the fact that we are trying to summarize the vice versa. However, the relationship may be more complex, and
relationship between two variables with a single number. some links may exist for only one ideological side but not the
Furthermore, the use of less known measurements may increase other. Indeed, right-wingers may be positive towards a particular
the difficulty of interpreting results, as many researchers would topic (e.g. country music), whereas left-wingers may have no
not be familiar with their meaning. defined opinion on the topic (i.e. some left-wingers may like it,
The use of these metrics to summarize relationships also has some may dislike it and some others may be neutral). An extreme
some important consequences for the interpretation of latent case of this is represented as an example in Fig. 2b. While the
constructs such as group identities. For example, let us suppose we interpretation of the correlation would be technically true that
run our analysis on 4 political items, including one item that “the prototypical right-winger, on average, likes country music
assessed peoples’ self-placement on a left-right continuum. Let us more than the prototypical left-winger,” it would still hide the
further assume that these procedures would provide a network like information that “right-wingers, in general, tend to have a clear
the “original network” shown in Fig. 2a. The results would lead us liking for country music, while left-wingers, in general, do not
to the following conclusion: as we move towards the left side of the have a clear preference.”
spectrum, respondents tend to answer more positively to both item Of course, many of these problems could be solved by
B and item C (we infer this from the positive correlation between B manually inspecting the relationship between pairs of items (i.e.
and C with the left-right self-placement item “LR”). Conversely, as their scatterplot). However, the number of relationships to
we move towards the political right, people become more and more explore is roughly N 2 =2 where N is the number of items. This

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 3

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Fig. 2 Exploration of response patterns using BNA. a Interpretation of a network from BNA. b Example of an asymmetric relationship. People on the
extreme right have a clear attitude stance, whereas people on the extreme left do not.

results in a very long and complex task, particularly when Introducing ResIN
considering that normal relationships are usually more complex As previously discussed, BNA usually considers items as nodes
than the extreme cases that we have simulated so far. and their correlation as the link’s weight. ResIN instead, simi-
Furthermore, inspecting each inter-item relationship in isolation larly to item-response theory, focuses on item-responses, and
would still not provide a good understanding of the overall consequently models item responses as nodes. For instance, if
attitude system. Indeed, the promise of BNA is to show key an item “vaccines are useful” has five levels, ranging from
properties of an entire attitude system within a single network “Strongly agree” to “Strongly disagree,” ResIN will produce 5
(ideally, at a single glance), while putting immediate attention on different nodes, each reflecting a different response option (e.g.
similarities and differences between groups. “Vaccines are useful: Strongly agree”, “Vaccines are useful:
Somewhat agree”, etc.). When dealing with a dataset, this
approach can be done through a process called “dummy cod-
A belief network with a latent space. In a recent publication ing.” This operation can be described as producing multiple
(Carpentras et al. 2022), we used a modified version of BNA and binary variables from a single variable (Jose, 2013). Keeping the
other methods to outline attitude dynamics in the context of previous example, we would obtain a binary variable for each
vaccination. The employed BNA modifications not only offer response option (e.g. “Vaccines are useful: Strongly agree”)
more detailed information about participants’ vaccine-related whose values would be 1 for every person selecting this item-
belief system, but one parameter extracted from it allowed us to response and 0 for every person selecting a different item-
make predictions of opinion change and behavior. This para- response. While some readers may think of dummy coding as a
meter was also strongly connected with spatial distance in the redundant process, we will see that it will allow us to have a
visualization, suggesting the idea of an underlying “opinion much more detailed description of the relationship between
space.” Comparably, in another manuscript (Lüders et al. 2024) items and, consequently, allow more flexibility in the attitude
we used the information about spatial locations in the visualiza- system. More specifically, ResIN abandons the assumption that
tion to predict political party identification. In the current work, survey responses have interval (or even ordinal) measurement
we explain and validate this variation of BNA as a spatial network properties at the individual level, and instead, it reconstructs
of attitudes (Barthélemy, 2011; O’Sullivan, 2014). In other words, spatial relations at social levels.
the output of this new method can be thought of as a network in Having defined what the nodes are, the next step is to identify
which each node holds a spatial coordinate in what we might call how to properly weight the links. Following the common
an attitude or identity space. Spatial networks have already been approach of BNA we calculate the weights as the correlation
used in situations such as networks of airports, where it is not between the corresponding node variables. Conveniently, as we
only important if two airports are connected by some flights, but are dealing with binary variables, we do not even have to choose
also their position and distance in the physical space. However, as between Pearson’s correlation, Spearman’s correlation, or Phi
we will discuss, in our case the x-coordinate will correspond to correlation (which was developed for calculating the correlation
positions in a latent space. To validate this procedure, we will of binary variables) (Guilford, 1941). Indeed, when dealing with
develop a mathematical model to have a better understanding of binary variables, all of these metrics produce exactly the same
how nodes are spatially located by the algorithm. Furthermore, result (Ekström, 2011). This is also very useful as many
we will compare the results that we obtain with our new method, researchers might be familiar with one (e.g. Pearson), but struggle
called ResIN, with results from Item Response Theory (IRT) to interpret values from different coefficients. Note that correla-
using both simulated and empirical data from political surveys. tions between responses of the same item (e.g. “Gun con-
Furthermore, we will show how ResIN can provide “in a glance,” trol:agree” and “Gun control:neutral”) will not be calculated. This
interesting results which are either hard or impossible to identify will be further justified later while discussing negative correla-
with the other methods. In doing so, we will demonstrate the tions, but we can immediately notice how these responses are
value of ResIN for studying belief systems as well as psychological mutually exclusive, thus not adding information to the system.
concepts like group identities. Also notice that, for two 5-levels items, ResIN will produce 25
In the next section, we will clarify how ResIN works and how it edge-weights (the correlations between all possible between-item
can be used to explore multiple belief systems within a single response pairs) while BNA will produce only one (i.e. the cor-
attitude space. Subsequently, we will develop a mathematical relation between the two items). This will be at the core of why
model to show how ResIN relates to IRT. Finally, we will validate ResIN can analyze more complex situations such as asymmetric
ResIN using simulated and real-world data. opinion spaces without losing information.

4 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

Fig. 3 ResIN and BNA visualization of a simulated dataset. a Pattern of simulated data obtained with the ResIN method and b pattern obtained from the
same data with classical BNA. Letters represent different items, while colors in a represent different levels (e.g. “strongly agree”, “weakly agree”, etc).

Up to this point, ResIN has produced an attitude network in The same can be understood as the possibility of guessing the
which nodes reflect dummy coded item-responses and links coffee preference of a person just by knowing their political
reflect the correlations between different nodes. However, as we affiliation. While our guess would be remarkably precise for
will see in the next step, ResIN also produces spatial information. Republicans and Democrats, this would not be the case for
Spatial networks are specific types of networks in which nodes independents. This can be easily visualized by ResIN as the cluster
have spatial coordinates in a (usually bi-dimensional) space. To of independent attitudes will not include any response about
obtain the spatial coordinates, ResIN uses a force-directed algo- coffee.
rithm (Fruchterman and Reingold, 1991). This algorithm models Figure 3b also depicts the pattern one would obtain using
every node similarly to an electrically charged particle (e.g. an classical BNA with the same data. In contrast to the results
electron) resulting in a repulsive force between them. This means obtained via ResIN, with classical BNA the information of
that nodes will try to maximize their distance with respect to each independents having no clear positioning on coffee would be lost.
other. At the same time, links act as springs pulling the connected The reason is that classical BNA produces only five nodes and 10
nodes together. Specifically, the bigger the link weights, the links, compared to the 25 nodes and roughly 100 links displayed
stronger the force pulling two nodes together. As a result, groups in ResIN.
of nodes that are strongly correlated will be positioned close to
each other, while nodes that are weakly correlated with the rest Connection to item-response theory
will appear further apart. One of the most important concepts in item-response theory
One may critically notice that the force-directed algorithm (IRT) is the so-called item characteristic curves (IC curves) (Van
allows ResIN to work only with positive edges, therefore forcing der Linden and Hambleton, 1997). Given a certain latent con-
us to neglect the negative correlations for this specific step. struct (e.g. left-right spectrum) and an item-response R, an IC
However, as we will see when exploring the mathematical model, curve f R ðθÞ represents the probability that a person with value θ
this will not result in a loss of precision, as negative correlations of the latent construct will select R as an answer (Van der Linden
contain mostly redundant information about the attitude system. and Hambleton, 1997; Thomas, 2011; Drasgow and Hulin, 1990).
Notice also that this does not mean that the information from the These curves are often characterized by several parameters such
negative edges is lost, but only that it is not used for determining as the pseudo-guess value, the discrimination, and the difficulty
the spatial location, while it can still be used for any other type of value. This represents both the strength of IRT, as it allows us to
analysis. finely tune items and scales, but also one of the main obstacles,
An example of our method is provided in Fig. 3a, where we resulting in a potentially overwhelming amount of information.
run the method on a simulated (i.e. “toy”) dataset. As we can As we demonstrated before, ResIN represents each item-response
see, we produced 5 items (labeled with letters from “a” to “e”), as a single node nR . ResIN therefore contains less information
each with 5 levels (identified by different colors), for a total of than IRT which makes it a poor candidate for precise scale
25 item-responses. Clusters (i.e. groups of nodes that appear tuning, but, as we will see, dramatically improves its applicability
close to each other) in this figure represent patterns of item- as a network tool.
responses which are frequently selected together by (simulated) As we will show in the following, the x-coordinate of each node
participants. For example, participants who selected one of the obtained through ResIN corresponds to the mean of the IC curve
red responses are also very likely to select the other red in IRT. To be able to perform calculations, we will have to make
responses (we see this from their spatial proximity). We can some stringent assumptions. However, in later sections, we will
also tell that a participant selecting a red response will probably re-test the equivalence of ResIN and IRT using simulations that
select a yellow response (from the high number of red-yellow are based on more relaxed assumptions. Finally, we will validate
links and their spatial proximity) but not a grey or blue the relationship between the two methods using empirical data.
response (from the relative lack of links and further distance to
the red cluster). Phi correlation and overlap. In this section, we will show that the
Notice also that all clusters have 5 nodes, besides the grey one. correlation between two item-responses (i.e. the weight of the link)
This means that while all five responses are important for char- provides information on “how close” two IC curves are (we will
acterizing the other clusters, only four responses characterize the formally define the concept of “closeness” in a few lines). Firstly, we
grey one. For example, this might represent Democrats (blue) can start by using the formula of the ϕ correlation coefficient for
having a strong correlation with “liking latte” and Republicans two columns x1 and x2 in a dataset as (Guilford, 1941):
(red) characterized by “disliking latte” (DellaPosta et al. 2015).
Therefore, we might expect independents (grey) then to be
 n n00  n10 n01
characterized by a neutral item-response (e.g. they will usually ϕ x1 ; x2 ¼ p11ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð1Þ
n1 n0 n1 n0
select: “neither like nor dislike”). However, we decided to simulate
a more realistic scenario in which some independents may like
latte, others may dislike it, and others may be neutral on the topic. Where nyz is the number of rows in which column x1 ¼ y and
Therefore, we observe no strong association between indepen- column x2 ¼ z. When one of the two entries is labeled with a dot,
dents and being neutral about latte. as in ny it means that that variable may be either 1 or 0. So n1 is

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 5

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Fig. 4 Correlation between items as function of item’s overlap and distance. a Relationship between correlation and the overlap of two curves.
b Correlation vs curves’ distance for Gaussian curves. Blue (dark grey) points represent correlation values below zero, while red (light grey) points
represent positive correlation.

counting only the number of rows in which x1 ¼ 1 independently a good measurement of the curves’ distance (being roughly
of the value of x2 . linear). However, for negative correlations, it progressively loses
To connect this formula to IRT, we can rewrite each one of the sensitivity (i.e. the derivative becomes smaller). Therefore, by
nyz variables as function of the IC curves. For example, for a dropping the negative correlations (as discussed in section 2) we
uniform population n1 can be written as: are losing very little information about the distance of the curves.
Z Alternatively, this could be also understood by imaging varying
n1 ¼ f 1 ðθÞdθ ð2Þ the distance Δ between two curves and observing how this affects
the correlation value. For positive correlations, this relationship
Because of this, we rewrite ϕ as: will be linear. For instance, moving from Δ ¼ 1 to Δ ¼ 1:5 the
correlation will decrease by 0.2, and similarly if we move from 1.5
I  b1 b2
ϕ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð3Þ to 2.0. Thus, a shift of 0.5 will roughly decrease the correlation by
b1 b2 ð1  b1 Þð1  b2 Þ 0.2. However, when the correlation is below 0, especially for Δ
Where I is the overlap of the two functions, defined as: bigger than 3, the relationship between the two variables changes.
Z Indeed, moving from 3 to 5, which is 4 times bigger than previous
I ¼ f 1 ðθÞf 2 ðθÞdθ ð4Þ shifts, will cause a decrease in correlation slightly bigger than 0.1.
Indeed, we can observe that close to 5 the curve is extremely flat.
And bi is the area of the IC curve i, written as: Practically this means that when curves are “close” (i.e. having
Z positive correlation) we can “observe” differences in position by
bi ¼ f i ðθÞdθ ð5Þ observing differences in the correlation value. However, this
information is progressively lost as the correlation becomes more
Something that can be noticed from Eq. (3) is that ϕ is linearly negative. Indeed, a correlation of −0.35 may represent a distance
dependent on the overlap I. Therefore, the bigger the overlap of 5 as well as 7. This means that for negative correlations the
between the two curves, the bigger the correlation value ϕ will be. process of estimating the spatial positioning from the correlation
Until now, we have not specified anything about the IC curves. value becomes rather unreliable.
Indeed, different theories from IRT use a different family of
curves, even if, in many cases, they are similar to Gaussian-like Forced-directed method and position. As previously outlined, to
functions. Therefore, for now, we will consider the curves to be obtain the final attitude space, we use a force-directed method to
Gaussian, obtaining the following overlap (Bromiley, 2003): estimate the position of each node. In such a configuration, each
Δ2
node experiences an attractive and a repulsive force following the
I ¼ K 1 eK 2 ð6Þ formulas (Fruchterman and Reingold, 1991):
Where Δ is the distance between the two curves calculated as the
2
F a ðd Þ ¼ w dk
difference between their mean (i.e. μ1  μ2 ) and K 1 and K 2
1
ð7Þ
k22
depends only on σ 1 and σ 2 . F r ðd Þ ¼  d
To make this clearer, in Fig. 4 we plotted the correlation values
(using Eq. 3) depending on the overlap I and the distance Δ of Where k1 and k2 are two constants, w is the link’s weight and d
two gaussian IC curves. As expected from the formulas, as the is the distance between the two nodes. In order to find the
overlap between the curves increases, the correlation increases position of an item response, we need to find the equilibrium of
linearly (Fig. 4a). Furthermore, the relationship between the the dynamic system as:
correlation and the distance between the curves (Fig. 4b) is ∑F i ðθÞ ¼ 0 ð8Þ
monotonic though not linear. Meaning that the slope becomes
smaller with increasing distance (i.e. flattening of the curve). The A non-approximated solution will be provided in the next
practical consequence of this is, that for positive correlations, ϕ is section, where we will run ResIN with simulated data. Here, to

6 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

ResIN is proportional to the position of the IC curves in item

response theory (IRT). Here, we will strongly relax the previous
assumptions and test the entire system all at once using simulated
data, showing an even more robust relationship between the two.
We hereby follow the below procedure:
1. We generated a series of curves using the graded model
from IRT (Samejima, 1969; Samejima, 2010).
2. We calculated the correlation among item-responses by
using Eq. 3 (i.e. from their IC curves) for a normally
distributed population, thus obtaining the network.
3. We use the force-directed method from the Python
Networkx package to estimate the position of the nodes
in the opinion space.
4. We use principal component analysis (Jollife and Cadima,
2016) to rotate the opinion space in such a way that the
main axis along which nodes are distributed will coincide
with the x-axis.
5. We compare the x-coordinate of each node with the mean
Fig. 5 IC curves’ mean versus position estimated by the force-directed of the corresponding IC curve from IRT.
method. We can notice how the relationship between the two variables is
Notice that, while many models in IRT suppose quite simple
monotonic and almost linear, besides the extremes where some
curves (e.g. all the same amplitude) (Van der Linden and
compression is present.
Hambleton, 1997), the graded model allows for a lot of
variation between curves. For example, in Fig. 6a we can see
provide more insight, we consider the special case in which all
the 5 curves for the 5 response-options of a single simulated
nodes but one are fixed at the coordinate of the mean of the
item. It is possible to see that IC curves vary in terms of
corresponding IC curve (i.e. θi ¼ μi ). Then, by supposing only
amplitude, standard deviation, and even shape. This com-
attractive forces, and approximating the correlation as propor- pletely relaxes the stringent requirements we imposed in
tional to the curve’s mean distance (as in Fig. 4b), we will
previous steps to be able to mathematically explore the
test whether the last node j will also be placed at the relationship between the two methods. Therefore, the graded
“correct” x-coordinate (i.e. μj ).
model allows for a test in much more complex (and realistic)
Therefore, to find the position θj;fin of the node at the situations.
equilibrium, we must solve the system:
In Table 1 we reported different values of Pearson’s correlation
  ϕij  2 k   2
  between the position estimated by the ResIN method and the
F θj;fin ¼ ∑ θj;fin  μi ¼ 3 ∑μj  μi  θj;fin  μi ¼ 0 mean value of the corresponding IC curve. For each configura-
i k1 k1 i
tion, we obtained a correlation coefficient of r ~ 0.95, meaning
ð9Þ that the position obtained from the ResIN method approximates
Where k3 is the constant of proportionality between ϕ and the very well the position of the curves from IRT.
difference of the means jμj  μi j.
We solved this equation numerically for different positions of Analysis of empirical data
the means. We randomly assigned all the μ values (including μj ) In the previous section, we tested our method against simulated
using a uniform distribution. Then we numerically found the θj;fin data which were obtained from item-response theory. This was
which minimizes the absolute value of the force calculated using instructive, as it allowed us to have a ground truth for checking if
Eq. 9. We repeated this process for 1,000 configurations (every the results were consistent with what we expected under different
time randomizing all the μ values). As depicted in Fig. 5, θj;fin is conditions. For a final validation, we will test whether we can
replicate this relationship with empirical data. In doing so, we will
proportional to μj for all the tested points. Indeed, we can also
also show that the ResIN method can produce interesting insights
calculate the correlation coefficient between the two, finding that would not be visible by inspecting the data with classical
r = 0.994. This suggests that by using ϕ as the link’s weight, the belief network analysis (BNA) and IRT.
force-directed algorithm should place the nodes close to the mean
value of the corresponding IC curve; at least, in this very Data collection and pre-processing. To perform this analysis, we
simplified condition. collected data from N = 402 Americans (Age: 18–81, M = 34.0,
Something that we can notice is that Fig. 5 presents some form SD = 11.6 Gender: male 203, female 196, non-binary 3) through
of “compression”, where the interval [0,1] is mapped into the the crowd working platform Prolific Academic. Participants
interval [0.2, 0.8]. This is due to the fact that we have neglected answered 8 items on political issues (Malka et al. 2014). Each item
the repulsive forces. Indeed, this approximation works particu- had 5 levels ranging from 1 = Strongly agree to 5 = Strongly
larly well for nodes that are close to the center, as their position is disagree. We then used these data to produce 8*5 = 40 item-
mainly determined by how they are “pulled” by the surrounding responses, calculated their correlations, and, finally, we used the
nodes. However, the position of the nodes at the periphery is force-directed method to produce the attitude network shown in
determined by the attractive and the repulsive forces as they will Fig. 7. The items were as follows:
experience no pull on one side. Notice, however, that this will not
be true for the full method, as it will include both attractive and 1. Abortion should be illegal.
repulsive forces. 2. The government should take steps to make incomes
more equal.
Simulations. In the previous sections, we have showed (using 3. All unauthorized immigrants should be sent back to their
very stringent assumptions) that the position of the nodes in home country.

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 7

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Fig. 6 Correspondance between ResIN and IRT for simulated data. a An example of the IC curves used for simulating one of the items. Notice how curves
have different amplitudes, standard deviations, and even shapes. b Position estimated by the ResIN method versus the mean of the corresponding IC curve
for the case of 8 items and 5 levels.

into Republican and Democrat was consistent with self-

Table 1 Correlation between the nodes and the mean of the identification, as we will explain in the next section.
corresponding IC curve.
Data analysis and insights. To confirm the relationship between
Items Levels |R| P value
ResIN and IRT we extracted the position of nodes using the
5 5 0.96 <10−14 ResIN method in Python. To calculate the mean values of the IC
10 5 0.95 <10−28 curves, we followed a completely independent approach. This was
20 5 0.96 <10−53
done to guarantee the independence of the two results that we
10 10 0.95 <10−53
5 10 0.95 <10−45
intend to compare. For the IRT analysis, we fed the data into the
graded model (Samejima, 1969; Samejima, 2010) from the ltm
package in R (Baker and Kim, 2004; Rizopoulos, 2007), from
which we extracted the IC curve parameters and means. Finally,
4. The federal budget for welfare programs should be we correlated the means of the curves with the nodes’ positions.
increased. This resulted in a correlation value of r = 0.97 (p < 10−27). The
5. Lesbian, gay, and trans couples should be allowed to relationship between the two results can be observed in Fig. 7b.
legally marry. This confirms that the x-axis position of nodes in the ResIN
6. The government should regulate businesses to protect the network maps very well to the mean of the corresponding IC
environment. curves in the used dataset. This observation confirms our
7. The federal government should make it more difficult to assumptions that the understanding of spatial distance in ResIN
buy a gun. (on the x-axis) can be interpreted as a distance on a latent
8. The federal government should make a concerted effort to variable.
improve social and economic conditions for African Now that the connection with a well-established theory (i.e.
Americans IRT) has been demonstrated, we can analyze Fig. 7a to see if it
Of these 8 items, 6 were written in such a way that the answer provides some useful insights. From visual exploration, we can
“Strongly agree” was more associated with a Democrat position and already notice that nodes of the same color (i.e. same levels on the
“Strongly disagree” with a Republican position. The remaining 2 left-right spectrum) are placed close to each other within the
items instead were inversely coded to prevent respondents from network. This observation is another confirmation that ResIN is
simply providing the same answer to all 8 items. After data “behaving” as expected, namely, that a person who selects a
collection, we inverted those two items, to obtain consistent specific item response (e.g. the most prototypical Republican
patterns. i.e. level 1 is the most conventionally Republican response response to item x) would select related responses across other
and level 5 is the most Democrat response for every item. Notice items as well.
that this is not needed for the main algorithm, but it is done only to Another confirmation that ResIN is working as expected is the
make visualization 7a clearer, as we use colors to visualize levels of fact that the blue cluster and the red cluster (i.e. the two most
agreement as follows: dark blue is used for dominantly Democrat- ideologically extreme) are placed at the two extremes of the
associated item-responses (level 1; strongly agree) and red for attitude space. Indeed, since ResIN produces results comparable
dominantly Republican-associated item-responses (level 5; strongly to IRT, we expect the pattern to be based on the left-right
disagree). Pale blue, grey, and orange are respectively levels 2 spectrum.
(weakly agree), 3 (neutral), and 4 (weakly disagree). If we did not A surprising result comes from the fragmentation that can be
reverse code the two items, their positions would still be the same, observed in the middle of Fig. 7a, where the system seems to be
but the colors would be inverted (e.g. red for level 1). split into two major clusters: one including the dark blue and part
We also collected information about self-identification as of the pale blue, and another bigger cluster including all the rest.
Republican or Democrat. Also in this case, we did not use this This visual cue can be confirmed by running Gephi’s modularity
information to obtain Fig. 7a, but only to confirm that its division algorithm (Lambiotte et al. 2008), based on the Louvain method

8 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

Fig. 7 Comparison between ResIN, IRT and BNA on empirical data. a Attitude network obtained by running the ResIN method on the data. Color has been
used to identify different levels. b Relationship between nodes position and the mean of the equivalent IC curve from IRT analysis. c Network obtained
using classical BNA on the same dataset.

Fig. 8 Analysis of the clusters obtained from empirical data. Coloring of the graph based on a clusters identified by Gephi’s modularity algorithm and
b self-identification (blue=democrats, red=republican).

(Blondel et al. 2008) which also identifies two main clusters. By exclusion, the other cluster should be the one containing
Nodes from the two clusters are colored respectively in orange Republican responses. Yet, it is important to notice that the
and violet in Fig. 8a. extracted Republican cluster contains also all the “neutral”
Since the links in our algorithm represent positive correla- responses and even three responses that could have been
tions, we can interpret each cluster as a response pattern which associated with Democrats (if one would naively follow an
is common in the population. For example, people selecting interval continuum in which “neutral” responses represent the
one of the responses of the left cluster are also very likely to qualitative midpoint, dividing Republican and Democrat opi-
select their other responses from the same cluster. Further- nions). Namely three of the pale blue nodes:
more, they are not very likely to select responses from the other
cluster. – Abortion should be illegal: somewhat disagree
As this cluster contains either strongly or weakly Democrat- – Lesbian, gay and trans couples should be allowed to legally
associated responses, it is easy to classify it as a Democrat cluster. marry: somewhat agree

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 9

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

– The government should regulate businesses to protect the example is the telescope, which was initially developed for ter-
environment: somewhat agree restrial purposes (i.e. looking at far objects, like ships) but then it
became a fundamental tool for the study of celestial objects. More
To make sure that this result is a real feature of the system and
recently, Ising’s models of atomic magnetic interaction have been
not an artifact we correlated each response with the self-
applied to the study of social influence (Ising, 1925) and even as a
identification variable (Fig. 8b). Here nodes’ color represents
basis for network psychometrics (Dalege et al. 2017). In a similar
the self-reported political identity of participants who most often
fashion, it is not impossible for an algorithm mimicking physical
selected that node as a response. For example, nodes colored in
forces (i.e. the force-directed algorithm) to produce results
blue are more likely to be selected by (self-identified) Democrats,
comparable to an established psychometric theory (IRT).
while nodes in red are more likely to be selected by (self-
In previous sections, we have already shown how the force-
identified) Republicans. This division replicates almost perfectly
directed model can produce high-quality results which are
the split in the responses provided by ResIN, confirming that the
extremely close to the ones from IRT (i.e. via mathematical
split produced by our method is not an artifact, but rather, it
modelling, simulations, and real data). Here we want to compare
reflects the fact that neutral as well as some moderate responses
this methodology with another popular approach: multidimen-
tend to be more frequently selected by Republicans than by
sional scaling (MDS).
Democrats.
In Fig. 10a, b we compare the 2D visualization of multi-
In the next section, we will offer a brief explanation of why we
dimensional scaling and ResIN for the same data simulated from
may observe such an asymmetrical pattern. Before doing so,
the graded model (following the same procedure as section
however, we will focus on the fact that ResIN was able to
“Simulations”) having 8 items, each with 5 levels. MDS was
recognize this unexpected split in the data without having access
calculated using the function MDS from sklearn.manifold in
to participants’ self-identification scores. Note that this informa-
python.
tion is undetectable in a network obtained from classical BNA
The first thing we can notice is that, obviously, classical MDS is
(Fig. 7c). As depicted in Fig. 9, even if the split between the two
missing the network structure, even if this could be solved by
clusters is embedded in the curves from IRT, the information
simply combining the information from the two methods. The
density in this figure is so high that recognizing this split simply
second noticeable result is that, even if the simulated data are
by looking at the curves is extremely challenging, if not
unidimensional, the visualization of MDS is curved, almost
impossible.
introducing an artificial second dimension. While this effect may
be surprising to some, it can be explained by the fact that IRT and
Further explanation of response-items’ positions in the MDS are different methods operationalizing dimensions in
attitude space. In our previous analysis we showed how different ways. Therefore, what is unidimensional in IRT can
responses that might naively be thought to be associated with correspond to multiple dimensions in other methodologies.
Democrat identity, such as “Lesbian, gay and trans couples should Despite this, it is interesting to notice how in ResIN the network
be allowed to legally marry: somewhat agree,” are in fact more mainly follows a unidimensional structure.
associated with Republican identity (i.e. the “somewhat agree” Another interesting comparison is which of the two methods
option was more often selected by Republicans than by Demo- (ResIN and MDS) produces results on the main axis which are
crats). A possible explanation for this anomaly could be that closer to IRT. Figure 10c shows the scatterplot that we produced
having clear (i.e. extreme) stances on abortion, gay rights, and in Fig. 7a where we compared the position of the nodes in ResIN
environmental protection is essential for the self-understanding with the mean of the IC curves when using real data. However, in
(i.e. identity) of what it means to be a Democrat. The same topics this case, we also added the scatterplot for MDS (orange) as well
might be less important for the average Republican voter which is as the dashed red line representing where points should be for
why Republicans can hold a wider spectrum of beliefs (note having a perfect agreement with IRT. By the naked eye we can see
however, that also for Republicans extreme responses seem to be that the blue points deviate less from the ideal curve. However, to
the most prototypical). avoid relying only on visualization, we also calculated the
We can understand this better by considering an example from respective correlations between each method and IRT. ResIN
a different context, such as the flat-Earth conspiracy theory. In produces a correlation of 0.976**** while MDS produces a
this case, when presented with an attitude measure such as “the correlation of 0.941****. Thus, both MDS and ResIN produce
Earth is not flat”, most people will select the most extreme option results extremely close to IRT in this application. One may even
(e.g. Strongly agree). Instead, people who are more connected to notice that ResIN’s correlation is larger than MDS’s one, but a
“alternative truths” (e.g. the Earth is flat, the Earth is a hologram, single test is not sufficient to conclude that ResIN’s results are
etc.) would be much more likely to select options such as closer to IRT.
“neutral” or “somewhat agree.” Put differently, showing any To test if really ResIN can outperform MDS, we produced
doubt at all would position you outside the mainstream. 100 simulations using the graded model. Each time we produced
Therefore, on a 5-level item, the pattern of answers will not be a new set of items and responses by randomizing the parameters
“2 levels which are selected by flat-earthers, 1 level selected by of the model. Even the total number of items was randomized
neutrals, and 2 levels selected by non-flat-earthers.” Instead, it each time between 5 and 10, and a number of levels was
would be: “4 levels selected by flat-earthers and 1 level selected by randomized between 3 and 6. For each simulation we compared
the non-flat-Earthers.” A similar pattern is also observed in the correlation between the x-coordinate of node’s position in
vaccine hesitancy (see Carpentras et al. 2022 for an extended each method and the mean of the equivalent IC curve. In 98% of
analysis). cases ResIN produced results closer to IRT than MDS did,
confirming the results previously observed with real data.
Further validation. Despite the robust relationship between The final validation we run confirms whether the force-
ResIN and IRT, skepticism may result from relying on the force- directed method is reproducible. This is an obvious concern, since
directed algorithm as, in the literature, it is mainly used for this type of algorithm randomizes the initial position of the
visualization purposes. However, it is crucial to notice that even if nodes, possibly obtaining different results each time it is run.
a tool has been mostly used for a specific purpose, it may still Although all the analyses presented so far have shown great
produce excellent results also in other contexts. A famous agreement between ResIN and IRT, which would not be possible

10 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

Fig. 9 IC curves obtained by using the analysis with the ltm package in R. IRT allows for a better exploration of the detail of each response-option.
However, the overall amount of information does not really allow to easily identify overall patterns as done in ResIN.

if the node’s positions were subject to extensive fluctuations, here the nodes and finally calculate the average correlations between
we want to better quantify the effect of the force-directed method them. For 50 simulations (which give us 1250 correlations) we
by keeping the network constant and re-running only the force- obtain an average correlation of 0.99985 with a standard
directed algorithm. For each run, we collect the x-coordinates of deviation of 2e-16. Three of these relationships are shown in

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 11

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Fig. 10 Comparison between ResIN and MDS. a Placing the nodes in the 2-dimensional space according to MDS. Notice how the structure is curved,
suggesting a second dimension even if IRT has used only 1. b The same data visualized with ResIN showing no curvature. c Scatterplot of the relationship
between ResIN (blue) and MDS (orange) compared to IRT. d Scatterplot showing the position of the nodes in ResIN when re-running the force-directed
algorithm. Closeness to the diagonal shows very minimal variations.

Fig. 10a. This shows that while the force-directed algorithm is Conclusion
stochastic, and can introduce some minor random fluctuations, In this paper, we validate ResIN as a robust methodological
its effect is very minimal. Furthermore, as previously discussed, approach for researchers who are studying attitude-related phe-
despite these fluctuations ResIN still produces closer results to nomena, particularly if they are interested in how specific
IRT than MDS. response options relate to each other in an identity-laden opinion
One may even ask if it would be possible to achieve such levels space. We developed ResIN to combine strengths from classical
of correlation using a chaotic method. While the intuitive answer BNA and item-response theory. While BNA offers researchers a
would be that this is impossible, here we want to confirm such simple tool to extract and depict connections between items,
results using some additional simulations. To confirm this point, item-response theory offers deeper insights into the ordinal (and
we have produced a method that places nodes in completely possibly non-symmetric) structure of items by considering the
random positions and correlates such positions with the mean of difference between different item responses. Our main aim with
the equivalent IC curve from IRT. Repeating this process 100 ResIN was to provide researchers with a new tool that allows for
times we obtain an average r-squared of 0.03 and an average the exploration of complex phenomena while still producing
p-value of 0.44. This clearly confirms that a chaotic method relatively simple and intuitive outputs.
would not be able to produce correlation values close to the one By testing ResIN on empirical data from the US electoral
observed in ResIN (i.e. above 0.95). context, we show that ResIN offers insight that neither BNA nor
To see this even better we re-run the chaotic algorithm one IRT could offer with similar ease. Indeed, ResIN allows inspection
million times to see how often the algorithm would produce of the responses separately for different groups (e.g. Republicans
correlations above 0.95. We found that in one million trials this and Democrats). While exposing less information than IRT,
case never happened. This is in perfect agreement with the p ResIN allows for the quick identification of structural response
values of the different validations of ResIN. Indeed, a p-value of patterns (such as the split between Republicans and Democrats)
10−28 (as we have found in previous sections) means that the which was not equally visible in either of the two other methods,
probability of getting such a result by pure chance is 1 in 10 and shows intuitively the non-linear and asymmetric relation-
octillions. These what-if scenarios show that if ResIN was chaotic, ships between identities and responses.
it would be impossible to obtain any of the results we have shown Since ResIN produces a spatial network, it offers both network-
in this article. like information as well as (latent) space information. The latter is

12 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x ARTICLE

related to IRT and reflects latent constructs. Specifically, a node’s Boutyline A, Vaisey S (2017) Belief network analysis: A relational approach to
position on the x-axis is representative of different values of the understanding the structure of attitudes. Am J Sociol 122(5):1371–1447
main latent construct (e.g. left-right spectrum). Thus, we can Brandt MJ, Sibley CG, Osborne D (2019) What is central to political belief system
networks? Pers Soc Psychol Bull 45(9):1352–1364
quantify how extreme each attitude is within our society. Bromiley P (2003) Products and convolutions of Gaussian probability density
The more network-based interpretation, instead, does not functions. Tina-Vis Memo 3(4):1
assume any latent construct and it relies simply on the idea of Burger J, Isvoranu AM, Lunansky G, Haslbeck J, Epskamp S, Hoekstra RH,
correlation. Attitudes which are strongly correlated (i.e. are Blanken TF (2022) Reporting standards for psychological network analyses in
usually selected together) will also appear close to each other and cross-sectional data. Psychol method 28(4)
Carpentras D, Lüders A, Quayle M (2022) Mapping the global opinion space to
be connected by a strong link. Instead, weakly related attitudes explain anti-vaccine attraction. Sci Rep 12(1):6188
will appear further from each other. Using this interpretation, we Chen Y, Li X, Liu J, Ying Z (2018) Robust measurement via a fused latent and
can see which attitudes cluster together, so forming a shared set of graphical item response theory model. Psychometrika 83:538–562
attitudes in the population. Similarly, we can see which attitudes Converse PE (1964) The nature of belief systems in mass publics. Crit Re
act as a bridge between two clusters, thus, eventually allowing 18(1-3):1–74
Dalege J, Borsboom D, Van Harreveld F, Van den Berg H, Conner M, Van der
dialogue between different social groups. Maas HL (2016) Toward a formalized account of attitudes: The Causal
Notice also how the algorithm does not require the data to be Attitude Network (CAN) model. Psychol Rev 123(1):2
ordinal, and the same method could be used, in principle, for Dalege J, Borsboom D, van Harreveld F, van der Maas HL (2017) Network analysis
studying also combinations of different data types, such as on attitudes: a brief tutorial. Soc Psychol Pers Sci 8(5):528–537
nominal and ordinal together. Future studies may also explore if De Winter JC, Gosling SD, Potter J (2016) Comparing the Pearson and Spearman
correlation coefficients across distributions and sample sizes: A tutorial using
such methodology could be reliably applied to multiple-choice simulations and empirical data. Psychol Methods 21(3):273
items (i.e. having non-mutually exclusive responses) and if the DellaPosta D (2020) Pluralistic collapse: The “oil spill” model of mass opinion
force-directed algorithm could be adapted to include information polarization. Am Sociol Rev 85(3):507–536
from the negative correlations. DellaPosta D, Shi Y, Macy M (2015) Why do liberals drink lattes? Am J Sociol
In sum, we believe that ResIN can be a useful and powerful tool 120(5):1473–1511
Drasgow F, Hulin CL (1990). Item response theory. In MD Dunnette & LM Hough
for social scientists interested in identity structures and (Eds.), Handbook of industrial and organizational psychology (pp. 577–636).
mechanisms underlying attitudes. Of course, we do not claim nor Consulting Psychologists Press
believe that this method should replace existing BNA methods, or Ekström J (2011). The phi-coefficient, the tetrachoric correlation coefficient, and
IRT. Instead, these three methods allow the exploration of the the Pearson-Yule Debate
attitude system at three different levels, just as a microscope, Epskamp S, Maris G, Waldorp LJ, Borsboom D (2018). Network psychometrics.
The Wiley handbook of psychometric testing: A multidisciplinary reference
telescope, and normal vision allow us to see the world at three on survey, scale and test development, 953–986
different levels; and each tool quickly reveals phenomena at a Epskamp S, Borsboom D, Fried EI (2018) Estimating psychological networks and
level of analysis that would be invisible or extremely complex to their accuracy: A tutorial paper. Behav Res Methods 50:195–212
find using only the other tools. Fruchterman TMJ, Reingold EM (1991). Graph Drawing by Force-directed Pla-
cement Software-Practice and Experiences, 21(11):1129–1164
Guilford JP (1941) The phi coefficient and chi square as indices of item validity.
Psychometrika 6(1):11–19
Data availability Haslbeck J, Waldorp LJ (2020) mgm: Estimating time-varying mixed graphical
Both data and codes used for this study are publicly available at models in high-dimensional data. J Stat Soft 93(8):1–46. https://doi.org/10.
the following link: https://github.com/just-a-normal-dino/AS22_ 18637/jss.v093.i08
Ising E (1925) Beitrag zur Theorie des Ferromagnetismus. Zeitschrift fur Physik
analysis_RESIN. 31:253–258. https://doi.org/10.1007/BF02980577
Isvoranu AM, Epskamp S, Waldorp L, Borsboom D (Eds.). (2022) Network psy-
Received: 12 May 2023; Accepted: 16 April 2024; chometrics with R: A guide for behavioral and social scientists. Routledge
Isvoranu AM, Epskamp S (2021). Which estimation method to choose in network
psychometrics? Deriving guidelines for applied researchers. Psychological
methods
Jollife IT, Cadima J (2016) Principal component analysis: A review and recent devel-
Note opments. Philos Trans Royal Soc A: Math Phys Eng Sci 374(2065):20150202
1 Notice that another network-based technique, called Network Psychometrics (NP), is Jose, PE (2013). Doing statistical mediation and moderation. Guilford Press
also present in the literature (Borsboom et al., 2021; Burger et al., 2022; Isvoranu et al., Kan KJ, de Jonge H, van der Maas HL, Levine SZ, Epskamp S (2020) How to
2022). NP has also been used for studying attitudes (Dalege et al., 2016; Dalege et al., compare psychometric factor and network models. J Intell 8(4):35
2017), categorical variables (Haslbeck and Waldorp, 2015; Isvoranu and Epskamp, Kertzer JD, Powers KE, Mintz A, Terris, L (2019). Foreign Policy Attitudes as
2021) and has included stability analysis (Epskamp et al., 2018) and comparisons with Networks. The Oxford Handbook of Behavioral Political Science
Item Response Theory (Marsman et al., 2018; Epskamp et al., 2018; Chen et al., 2018). Klein O, Spears R, Reicher S (2007) Social identity performance: Extending the
However, it is crucial to notice that NP mostly focuses on using partial correlations strategic side of SIDE. Pers Social Psychol Rev 11(1):28–45
and regularization methods, thus attempting to remove the contribution of other Krempel L (2011). Network visualization. The SAGE handbook of social network
variables from the links. This is very different from the approach of BNA, where the analysis, 558–577
focus is on using the correlation between variables to describe the overall social-level
Lambiotte R, Delvenne JC, Barahona M (2008). Laplacian dynamics and multiscale
modular structure in networks. arXiv preprint arXiv:0812.1770
information structure. ResIN shares the same approach and goal of BNA and, because
Lüders A, Quayle M, Maher P, Bliuc A-M, MacCarron P (2024) Researching
of this, in the article, we will not focus on NP nor on comparisons with it.
attitude–identity dynamics to understand social conflict and change. Eur J
Soc Psychol 00:1–8. https://doi.org/10.1002/ejsp.3022
References Lüders A, Dinkelberg A, Quayle M (2022) Becoming “us” in digital spaces: How
Baker FB, Kim SH (2004). Item response theory: Parameter estimation techniques. online users creatively and strategically exploit social media affordances to
CRC press build up social identity. Acta Psychol 228:103643
Barthélemy M (2011) Spatial networks. Phys Rep 499(1-3):1–101 Lüders A, Carpentras D, Quayle M (2023) Attitude networks as intergroup reali-
Bliuc AM, McGarty C, Thomas EF, Lala G, Berndsen M, Misajon R (2015) Public ties: Using network-modelling to research attitude-identity relationships in
division about climate change rooted in conflicting socio-political identities. polarized political contexts. Br J Socl Psychol, 63(1):37–51
Nat Clim Change 5(3):226–229 Marsman M, Borsboom D, Kruis J, Epskamp S, van Bork RV, Waldorp LJ, Maris G
Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E (2008) Fast unfolding of (2018) An introduction to network psychometrics: Relating Ising network
communities in large networks. J Stat Mech: Theory Exp 2008(10):P10008 models to item response theory models. Multivar Behav Res 53(1):15–35
Borsboom D, Deserno MK, Rhemtulla M, Epskamp S, Fried EI, McNally RJ, Maher PJ, MacCarron P, Quayle M (2020) Mapping public health responses with
Waldorp LJ (2021) Network analysis of multivariate data in psychological attitude networks: the emergence of opinion‐based groups in the UK’s early
science. Nat Rev Methods Primers 1(1):58 COVID‐19 response phase. Br J Soc Psychol 59(3):641–652

HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x 13

ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-024-03037-x

Malka A, Soto CJ, Inzlicht M, Lelkes Y (2014) Do needs for security and certainty Funding
predict cultural and economic conservatism? A cross-national analysis. J Pers Open access funding provided by Swiss Federal Institute of Technology Zurich.
Soc Psychol 106(6):1031
McGarty C, Thomas EF, Lala G, Smith LG, Bliuc AM (2014) New technologies,
new identities, and the growth of mass opposition in the Arab Spring. Polit
Competing interests
The authors declare no competing interests.
Psychol 35(6):725–740
O’Sullivan D (2014). Spatial network analysis. In Handbook of regional science
(pp. 1253–1273). Springer, Berlin, Heidelberg Ethical approval
Postmes T, Haslam SA, Swaab RI (2005) Social influence in small groups: An Data collection was performed in line with the principles of the Declaration of Helsinki
interactive model of social identity formation. Eur Rew Soci Psychol 16(1):1–42 and received approval from the Research Ethics Committees of the University of Lim-
Quayle M (2020). A performative network theory of attitudes. PsyArXiv. https:// erick, 2020_02_03_EHS.
doi.org/10.31234/osf.io/mh4z8
Rizopoulos D (2007) ltm: An R package for latent variable modeling and item
response analysis. J Stat Softw 17:1–25 Informed consent
Samejima F (1969). Estimation of latent ability using a response pattern of graded Informed consent was obtained from all participants.
scores. Psychometrika monograph supplement
Samejima F (2010). The general graded response model. In ML Nering & R Ostini Additional information
(Eds.), Handbook of polytomous item response theory models (pp. 77–107). Correspondence and requests for materials should be addressed to Dino Carpentras.
Routledge/Taylor & Francis Group
Serrat O (2017) Social network analysis. In Knowledge Solutions. Springer, Reprints and permission information is available at http://www.nature.com/reprints
Singapore
Thomas ML (2011) The value of item response theory in clinical assessment: a Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
review. Assessment 18(3):291–307 published maps and institutional affiliations.
Van der Linden WJ, Hambleton RK (1997) Handbook of item response theory.
Taylor & Francis Group 1(7):8
Wakita K, Tsurumi T (2007). Finding community structure in mega-scale social
Networks. In Proceedings of the 16th International Conference on World
Wide Web (pp. 1275–1276) Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
Acknowledgements appropriate credit to the original author(s) and the source, provide a link to the Creative
This project has received funding from the European Union’s Horizon 2020 Research Commons licence, and indicate if changes were made. The images or other third party
and Innovation Programme under the Marie Skłodowska-Curie grant agreement No material in this article are included in the article’s Creative Commons licence, unless
891347 and from the European Research Council (ERC) under the European Union’s indicated otherwise in a credit line to the material. If material is not included in the
Horizon 2020 Research and Innovation Programme (Grant agreement No. 802421). article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://creativecommons.org/
Author contributions licenses/by/4.0/.
DC, AL, and MQ developed the idea and wrote all the versions of the article. Addi-
tionally, AL collected the experimental data, MQ supervised the project and DC devel-
oped all the analysis and simulations. © The Author(s) 2024

14 HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | (2024)11:589 | https://doi.org/10.1057/s41599-024-03037-x