Alyssa Jongeneel, 26th October 2020

PhD-student at Vrije Universiteit Amsterdam (department of Clinical Psychology), psychologist and product manager psychosis at Parnassia


In the Netherlands, we conducted a large national study to test the effects of the Temstem app on voice-hearing in daily life. Temstem was developed years ago as a tool that voice-hearers can use any time of the day to gain more control over their voices (for study protocol and more information about how the app exactly works, see Jongeneel et al., 2018). The app can be used in addition to other treatment (such as anti-psychotics or CBT), and is also usable stand-alone. Developing an app like Temstem has the advantage that persons not only get help while they see a therapist, but also when they’re home, struggling with their voices on their own. Almost everyone has a smartphone nowadays and previous studies have shown that persons who hear voices, are interested in using apps to increase their mental health.

During our study, we asked participants to use, besides Temstem, another (Experience Sampling Method; ESM) app. The app asked (i.e. ‘beeped’) ten times a day, for 6 days in a row, several questions about how participants felt emotionally, whether they heard voices, what they were doing with whom etcetera. This way, we generated a wealth of valuable daily-life data. 

For this study (for publication, see Jongeneel et al., 2020), we analyzed the ESM data of 95 voice-hearing individuals by network analysis. Network analysis is a visualisation of several multiple regression analyses and has the potential to clarify how certain factors are connected over time. Since voice-hearing often fluctuates during the day, we were interested in the factors that contribute to these fluctuations. This enabled us to possibly provide valuable insights into potential mechanisms and targets for voice-hearing interventions.

Since network analysis is a rather new technique and some find it difficult to understand, I will give a short explanation about the different types of networks. 


Relations between voice-hearing and negative affect, positive affect, uncontrollable thoughts, dissociation, and paranoia were analysed in three types of networks: contemporaneous, between subjects, and temporal networks. 

The contemporaneous model (see Figure 1) includes all separate data points (i.e. filled in questionnaires) per variable. It analyses unique, concurrent relations between the variables. For example: at Monday 09.00 AM person X heard voices and was at the same time experiencing negative emotional feelings. This network indicates mainly how factors are related on the (very) short term.

Figure 1. Contemporaneous network.

The between subjects network (see Figure 2) mostly reflects the more common cross-sectional analysis. It includes the participants’ mean of each variable during this week. So, if a participant filled in 40 questionnaires during 6 days, the 40 answers to the question: “do you hear voices?” are totalled and divided through 40 to calculate the mean. An example might be: during the week, in person X voice-hearing is related to paranoid feelings. So, in general, when person X hears voices, he is also paranoid. This network indicates how factors are generally, on a longer term, related. Both contemporaneous and between subjects networks do not provide any indication of causality.

Figure 2. Between subjects network. 

Finally, the temporal network (see Figure 3) uses data with a time lag (in this study, about 90 minutes), to predict for example whether voice-hearing  at a certain timepoint (Monday 09.00 AM) is preceded by dissociation at the previous timepoint (Monday 07.30 AM). Although there is a lot of debate about whether this model indicates causality (A leads to B), it at least shows whether factor A precedes B, which is one of the requirements to indicate causality.

Figure 3. Temporal network.

Results and Discussion

The network visualisations can be found in our published paper (Jongeneel et al., 2020). Results showed that voice-hearing occurs together with the other mental state variables, but it seems to be only connected to uncontrollable thoughts on the longer term. This is in contrast to previous research and this might be explained by our use of partial, i.e. unique, correlations. Also, we could not replicate previous findings that voice-hearing is predicted by any of the factors. This might be due to our analytic approach (multilevel network analysis versus single regression analyses as previous studies used), or our study sample; the participants in our study were severely ill and therefore potentially beyond the ‘tipping point’ of fluctuating clinical symptoms to a more chronic state whereby symptoms mainly maintain themselves. 

We interpret the results of the three networks altogether as following. A stimulus (for example a stressful event or negative thought) activates a factor that activates all other factors in a very short time period. Once activated, voice-hearing maintains itself without input from a stimulus. If this theory appears to be right, it is very important to assess whether voice-hearing is still reactive to triggers, so you can intervene on these triggers, or not so you have to breach the vicious circle by for instance challenging the negative content of the voices by Cognitive Behavior Therapy. 

We think it would be very interesting for future research to investigate the dynamics of voice-hearing in other study samples, for example individuals with a first episode psychosis. Also, individual networks might be valuable to analyse. And the networks might differ between subtypes of voices (e.g. commanding, threatening) which would be important to map.

Despite of the study limitations (see publication), we believe results of analysing (individualised) daily life data by network analysis are a valuable addition to our current knowledge about voice-hearing.


Jongeneel, A., Aalbers, G., Bell, I., Fried, E. I., Delespaul, P., Riper, H., … Berg, D. van den. (2020). A time-series network approach to auditory verbal hallucinations: Examining dynamic interactions using experience sampling methodology. Schizophrenia Research215, 148–156.

Jongeneel, A., Scheffers, D., Tromp, N., Nuij, C., Delespaul, P., Riper, H., … Van den Berg, D. (2018). Reducing distress and improving social functioning in daily life in people with auditory verbal hallucinations: study protocol for the ‘Temstem’ randomised controlled trial. BMJ Open8, e020537.