Streamlining of the Mellow Parenting Observation System for research and non-specialist clinical use

04 October 2019

Researchers Lucy Thompson, Gerry King, Eleanor Taylor, Christine Puckering and Phil Wilson explored whether a streamlined version of a tool to assess parent-child interactions provided comparable data to the full system and gathered feedback from health visitors.

Authors: Lucy Thompson, Senior Research Fellow, Institute of Applied Health Science, University of Aberdeen, Gerry King, Research Nurse, Institute of Applied Health Science, University of Aberdeen, Eleanor Taylor, Undergraduate Student, Institute of Applied Health Science, University of Aberdeen, Christine Puckering, Retired Clinical Psychologist, Mellow Parenting, Glasgow and Phil Wilson, Professor of Primary Care and Rural Health, Institute of Applied Health Science, University of Aberdeen.

Email: [email protected]

Research summary

  • Researchers took the Mellow Parenting Observation System (MPOS), a promising but complex tool that can be difficult and time-consuming for resarchers to use reliably, and developed a streamlined version – Child and Adult Relationship Observation (CARO) – which includes similar information and may be more accessible to practitioners.
  • They then applied the streamlined tool to data generated from a previous study using MPOS, and compared the results, finding a strong positive correlation between analyses of the same videos using MPOS and CARO.
  • The team went on to design a training package to deliver to health visitors, who are at the forefront of observing and assessing parent-infant interaction, but lack a formal, standardised system.
  • They invited a group of experienced practitioners to take part in training and provide feedback. The practitioners gave CARO positive feedback and said they would like to see it used in daily practice.
  • The study suggests that CARO would not sacrifice any of the quantitative power that the more complex MPOS system provided.
  • CARO shows promise as a simple observational tool without technical aids for use in clinical practice. Development work is ongoing to ensure it is user-friendly, while maintaining its reliability as a research measure.


Parent and infant interaction is very fast, fluid and variable. There are observation tools for practitioners and researchers, but there remains a need for a reliable and valid tool that can be used in everyday practice in the home or clinic. This paper describes the development of a streamlined version of the Mellow Parenting Observation System (MPOS): the Child and Adult Relationship Observation (CARO). MPOS has proven predictive validity, but is time consuming and demanding to learn, to achieve research reliability, and to analyse. CARO has been developed to make the best of the qualities of MPOS, but in a more streamlined and user-friendly way. We found a strong positive correlation between analyses of the same videos using MPOS and CARO. CARO shows promise as a simple observational tool for use in clinical practice. Development work is ongoing to ensure that CARO is user-friendly, while maintaining its reliability as a research measure. 

The Mellow Parenting Observation System (MPOS) was developed as a clinical, forensic and research tool to describe the interaction observed during video-recorded caretaking, with a specific focus on the interaction quality, rather than on individual parental or child behaviours. It involves trained observers analysing interactions on six dimensions (Table 1) with both positive and negative components. Thus far MPOS has primarily been used as a research tool and as a platform for strength-based video feedback to parents and carers, but systematic observation of parent-child interaction could be a useful tool for health visitors and other practitioners working with families.

The quality of parent-child relationships correlates significantly with lower rates of mental and physical ill health in childhood and adulthood (Latimer et al., 2012; Puckering et al., 2014). In the UK, health visitors incorporate assessment of the parent-infant relationship as part of their routine monitoring of child wellbeing and development. Despite this, the methods of evaluation and the training available for health visitors specifically on assessing parent-infant relationship quality are not currently regulated or standardised (Kristensen et al., 2017). Health visitors must rely, therefore, on a sound knowledge of known risk factors, direct observation of behaviours and intuitive reasoning to evaluate the parent-infant relationship, as well as drawing on personal experiences (Wilson et al., 2008). The lack of a formal and widely accepted assessment tool is especially challenging for newly qualified health visitors, who lack a wide repertoire of clinical experiences (Kristensen et al., 2017).

There is a range of methods available to observe and evaluate the quality of parent-infant interactions, including detailed complex methods such as the CARE-Index (Crittenden, 1998), the PIRGAS (Zero to Three/National Center for Clinical Infant Programs, 1994), and the PIRAT (Broughton, 2014), as well as simpler methods such as the PIIOS (Svanberg et al., 2013). All the systems have their strengths and adherents, as well as some deficits (see Box 1 for a non-exhaustive comparison of these tools). The CARE-Index has proven very difficult for users to learn and become reliable. The PIRGAS is theoretically related to the Zero to Three multiaxial diagnostic system and relies on clinical rating which is not standardised, but rather based on a clinician’s individual judgement. This makes it a less subtle tool than a behaviourally based systematic observation. Similarly, PIRAT divides ratings into being of no concern, some concern or significant concern with no underlying behavioural basis defined. The PIIOS, explicitly developed with the primary care setting in mind and showing good reliability for high-risk dyads, focuses on babies only up to around 7 months of age. A gap therefore exists for a clinically useful, behaviourally based measure that can be used throughout the pre-school years and can be validated psychometrically.


Age range


Ease of use


0 – 72 months

Clinical impression

Intense training

Difficult to become reliable

PIRGAS – Parent-Infant Relationship Global Assessment

0 – 36 months

Clinical impression – not standardised

Intense training

Requires clinical interview as well as video observation

PIRAT – Parent-Infant Relational Assessment Tool

0 – 24 months

Clinical impression

Intense training

Designed for clinicians and researchers

PIIOS - Parent-Infant Interaction Observation Scale

2 – 7 months

Clinical impression

Intense training

Designed for primary care practitioners

MPOS – Mellow Parenting Observation System

0 – 60 months

Count of observed behaviours

Intense training

Designed for researchers or clinicians

Box 1: Examples of available observation measures


The Mellow Parenting Observation System (MPOS) was developed as a clinical, forensic and research tool to describe the interaction observed during video-recorded caretaking, with a specific focus on the interaction quality,  rather than on separate parental or child behaviours which are generally the basis of other methods. Using MPOS, the observer records every instance of certain key interactional behaviours usually seen in a normal care-taking routine, such as a meal or playtime. The quality of interaction can be best assessed if the scenario being observed includes the parent having to negotiate an agenda (e.g. a nappy change or meal time) as this allows an assessment of interaction quality when the parent is aiming to gain the child’s cooperation; more reflective of everyday interaction than play which should be entirely child-led. MPOS samples six dimensions (Anticipation, Autonomy, Responsiveness, Co-operation, Distress and containment, and Control), each of which has positive and negative components (see Table 1).   For example, a positive behaviour in the Anticipation dimension might be a parent playing ‘trains’ to encourage their baby to open their mouth for the next spoonful of food. Positives and negatives are scored separately and have been shown to be statistically independent (Thomson et al., 2014). Reviews of independent video analysis showed there to be reasonable agreement between different observers for overall positive and negative behaviours, however, agreement within the separate dimensions was less good (suggesting that while observers find it relatively easy to agree that a behaviour is either positive or negative, it is more difficult to agree in which dimension it fits, particularly when multiple dimensions may be operating simultaneously (Puckering et al., 2014). 

MPOS has also shown promise as a predictive tool (using overall positive / negative interactions). Previous research using banked video data of one-year-olds interacting with their parent / caregiver has shown that each increase of one positive parental behaviour per minute of observed interaction predicted 15% lower odds of a child later (age 7 years) receiving a disruptive behaviour disorder diagnosis (Puckering et al., 2014). It has also been shown that systematic analysis of videotaped interactions tells us more about a child’s risk of later problems than could be learned just from looking at known risk factors (such as maternal education, child’s gender, smoking and alcohol use in pregnancy, and prenatal depression) (Thompson et al., 2014).

As well as showing promise as a research tool, the principles of MPOS (i.e. application of the same dimensions in observing video-recorded interaction) have been successfully applied in a clinical context (Mellow Parenting programmes) as part of a strength-based parent-led feedback discussion (Puckering, McIntosh, Hickey, & Longford, 2010).

Although MPOS shows promising potential, it has similar limitations as other detailed observation tools in that it is complex, and it can be difficult and time-consuming for researchers to become reliable.  For example, a single behaviour can be scored in more than one dimension simultaneously; a ‘good enough’ user might count this interaction as positive, but may fail to note all the dimensions in which it could be counted. For example, “Just one more bite and you will have finished your dinner” in a positive tone of voice would receive a positive tally on the tone aspect of the responsiveness dimension and on the anticipation dimension.






  1. Co-operation
  1. Co-operation


  1. Control



  1. Anticipation



  1. Autonomy
  1. Autonomy



  1. Responsiveness
  1. Responsiveness


  1. Distress


Available codes

Multiple per interaction

Single per interaction


Total positives / negatives (no potential maximum)

Can convert to ‘rate per minute’

Set parameters (max 6 positive / negative per minute)

Can convert to ‘rate per minute’

Extra ratings

Can also rate level of involvement for each 10-second segment


Table 1: Comparison of key features of MPOS and CARO


The complexity in MPOS also makes it difficult to compare sets of interactions directly, as the range of possible positive and negative total scores is (technically) limitless. Simplifying the analysis so that counting an interaction as simply positive, negative, or neither, in research could help to standardise the scoring. It is also potentially useful to limit the number of possible codes applicable within a given time segment (in this case, 10 seconds). Thus, in a 5 minute segment of interaction, the range of potential ‘positive’ or ‘negative’ rates will each be fixed at 30 (i.e., each minute contains 6 x 10 second segments). A further limitation of the MPOS is that analysis is a time-consuming process that requires significant training and an iterative post-training accreditation process. Consequently, it has only been used reliably for research purposes by psychologists; it has been difficult for non-psychologists to gain research reliability in the system.

In the present study we aimed to explore whether a streamlined version provides comparable data to using the full MPOS system, and to obtain feedback from a small group of experienced health visitors. We developed the simplified coding system, newly named the Child and Adult Relationship Observation (CARO). It is fundamentally different from MPOS in three ways (see Table 1). First, it counts only a maximum of one positive and one negative instance of interactional behaviour in each 10-second interval. Second, the six dimensions of the previous system have been reduced to three: co-operation, autonomy and responsiveness. Finally, it removes the potential for multiple coding of each interaction element. MPOS also includes the possibility of rating the caregiver’s level of involvement in the interaction – this has been removed from CARO.


We applied the simplified system - CARO - to data generated from a previous study using MPOS (n=55). For every 10-second interval where, in the old system, there may be several instances of observed interactional behaviour, we allotted a maximum of one positive and one negative per 10-second period. The old and new counts were then entered into a statistical analysis programme (IBM Corp., 2015) and correlations run between the rates of positive and negative behaviours per minute. We used Spearman’s rank order correlation coefficient to compare the rates between the old MPOS and the new CARO observations. Scatter plots were produced for visual comparison (see Figure 1).


We found good correspondence between ratings on MPOS and CARO using the positive and negative rates per minute (positive: r=.97; p<.0001; negative: r=.98; p<.0001). These strong positive correlations indicate that reducing the ratings to a maximum of one positive and / or negative behaviour every ten seconds does not reduce the accuracy of the measure, compared to MPOS. The comparison using rates per minute was considered more useful as it dealt with outliers caused by varying lengths of video recording (see Fig 1 below).

In order to begin examining the practical utility of CARO, we went on to design a training package to deliver to health visitors and invited a group of experienced practitioners to take part in training and provide feedback. Their views can be summarised as follows:

Providing a common language: One practitioner had used the observation system as a checklist in assessing a mother / toddler relationship. She reported that the system provided a language that strengthened the evidence for making a referral to another agency. In this case using the system facilitated a successful referral to psychological services, which had previously proven difficult for health visitors to access for families.

Learning important observation skills: Practitioners would like to see more of this sort of training as standard. There is a perception that, since the implementation of national policies designed to improve inter-agency communication in the interests of children, there is less focus on child development or infant mental health in training, which is seen as far from ideal. Recent implementation of training in the Solihull approach (Douglas & Brennan, 2004) is welcome, but practitioners felt that CARO would be easier to apply in everyday practice.

Security of information: Practitioners had concerns about how to approach the issue of security and trust around video recording interactions. If video recording was to be done in clinical practice, there were concerns as to what/whose device should/could be used, data ownership, and how to reassure parents regarding confidentiality.  This section offers a good application to practice which the readers will benefit from.

Practicalities: Practitioners need to be using CARO / practicing CARO regularly to stay skilled. The reality is that health visitors do not have the available time to be making and coding videos on a regular basis. A method of applying the concepts in real-time (i.e., without video) rating would be welcomed. Practitioners also suggested that a quick-reference guide, such as a laminated A5 card showing a checklist, would be useful.


Analysis of video-taped parent-infant interactions using the MPOS and the simplified version, CARO, are comparable. This suggests it is reasonable to apply CARO as a research tool without sacrificing any of the quantitative power that a more complex system, in this case MPOS, could provide. Whether there is a loss of richness of data in a qualitative sense will need to be investigated with further development of the tool and validation against not only MPOS but other established measures of parent-infant interaction quality.

Initially, we anticipated the benefits would be realised predominantly in making the system more accessible for research purposes. However, in addition to this, we believe that it can be more easily used by non-specialist clinicians in their day-to-day practice. Practitioners gave positive feedback about using it in practice, and were keen to see more of this sort of training in their work. There is a small but limited literature describing simple observation systems, (e.g. Svanberg et al., 2013), but more evidence to support their use in real time (i.e. not dependent on video recording) by non-specialist practitioners is needed. CARO, in utilising core psychological concepts about the quality of the relationship (e.g. reciprocity) rather than relying simply on behaviour of either carer or child, may mean that it can provide practitioners not only with a tool for observation, but with a conceptual foundation and specific language to communicate about relationships both with families and with fellow practitioners, including in written communication for onward referral to more specialist services. Information from the observations can be used to complement all the existing information already gathered about families, and to inform decisions about care and support.

Having simplified the scoring system, we have developed a smartphone app to make recording of interactions more accessible. This app is currently in a development / testing phase with a view to it being accessible to practitioners in the field. It is web-based (rather than a stand-alone app), so with internet connectivity could be accessed via standard health service and other IT systems. The next steps for the research include developing training for practitioners (health visitors and general practitioners, initially) and collecting data using the smartphone app. This will allow us to examine how user-friendly CARO and the associated app are, and the reliability of the system, both with and without using the app.

We anticipate that the CARO will prove useful in the day-to-day work of community nurses as a triage tool without technical aids.  Where difficulties in the parent-child relationship are apparent, CARO could offer a triage tool indicating the need for further assessment, a mechanism for identifying families that could benefit from intervention, and for communicating the level of concern to colleagues and other agencies. 

Declarations of interest: CP was until recently Programme Director of Mellow Parenting, a charity registered in Scotland who hold the copyright of the Mellow Parenting Observation System.

Acknowledgements: Thanks to the parents who allowed their interactions with their children to be recorded and used for research purposes.

Funding: This research was partially funded by the Grant Foundation through the Development Trust of the University of Aberdeen.


Broughton C. (2014) Measuring parent–infant interaction: the Parent−Infant Relational Assessment Tool (PIRAT). Journal of Child Psychotherapy 40(3): 254-270. See (accessed 12 September 2019).

Crittenden PM. (1998) CARE-Index Manual (3rd revision). Miami, FL: Unpublished manual.

Douglas H & Brennan A. (2004) Containment, reciprocity and behaviour management: Preliminary evaluation of a brief early intervention (the solihull approach) for families with infants and young children. Infant Observation 7(1): 89-107. See (accessed 12 September 2019).

IBM Corp. (Released 2015) IBM SPSS Statistics for Windows, Version 23.0. Armonk, NY: IBM Corp.

Kristensen IH, Trillingsgaard T, Simonsen M & Kronborg H. (2017). Are health visitors’ observations of early parent–infant interactions reliable? A cross-sectional design. Infant Mental Health Journal 38(2): 276-288. See (accessed 12 September 2019).

Latimer K, Wilson P, Kemp J, Thompson L, Sim F, Gillberg C, Puckering C, Minnis H. (2012) Disruptive behaviour disorders: a systematic review of environmental antenatal and early years risk factors. Child Care Health and Development 38(5): 611-628. See (accessed 12 September 2019).

Puckering C, Allely CS, Doolin O, Purves D, McConnachie A, Johnson PC, Marwick H, Heron J, Golding J, Gillberg C, Wilson P. (2014) Association between parent-infant interactions in infancy and disruptive behaviour disorders at age seven: a nested, case-control ALSPAC study. BMC Pediatrics 14, 223. See (accessed 12 September 2019).

Puckering C, McIntosh E, Hickey A, Longford J. (2010) Mellow Babies: a group intervention for infants and mothers experiencing postnatal depression. Counselling Psychology Review 25: 28-40. See (accessed 12 September 2019).

Svanberg PO, Barlow J, Tigbe W. (2013) The effectiveness of training in the Parent-Infant Interaction Observation Scale for health visitors. Journal of Health Visiting 1(3): 162-166. See (accessed 12 September 2019).

Thompson L, Allely C, Puckering C, Purves D, Marwick H, McConnachie A, Johnson PC, Golding J, Gillberg C, Wilson P. (2014) P3-87 A Testable model for the prediction of psychiatric disorder at seven years. Paper presented at the World Association for Infant Mental Health 14th World Congress, Edinburgh, UK.

Thomson RM, Allely CS, Purves D, Puckering C, McConnachie A, Johnson PC, Golding J, Gillberg C, Wilson P. (2014) Predictors of positive and negative parenting behaviours: evidence from the ALSPAC cohort. BMC Pediatrics, 14, 247. See (accessed 12 September 2019).

Wilson P, Barbour RS, Graham C, Currie M, Puckering C, Minnis, H. (2008) Health visitors’ assessments of parent–child relationships: A focus group study. International Journal of Nursing Studies 45(8): 1137-1147. See (accessed 12 September 2019).

Zero to Three/National Center for Clinical Infant Programs. (1994) Diagnostic Classification of Mental Health and Developmental Disorders of Infancy and Early Childhood (PIRGAS). Arlington: Zero to Three/National Center for Clinical Infant Programs.