Taking a whole-of-system approach to food packaging reduction

Using network analysis to identify leverage points based on causal loop diagrams leads to false inference

“…The problems with the use of betweenness- and closeness centrality to identify “structurally important nodes” 66 are far from unfamiliar, where many of the problems previously detected in other fields 61 also apply to CLDs and several papers applying network analysis metrics to CLDs also hint at potential limitations 8 , 23 , 24 , 26 , 28 , 36 – 38 , 42 , 52 . It has already been shown for different types of networks that betweenness- and closeness centrality do not provide reliable results over networks that differ slightly in structure, for example when a node is included or excluded 61 , 69 , 70 .…”

“…Challenges regarding the specification of boundaries-i.e., which nodes should be included in and excluded from the network-which critically affects betweenness-and closeness centrality results, have accordingly also been extensively discussed in other fields 69,74,75 . This and other bodies of literature do not seem to have been considered in research using CLDs, presumably because, even though network analysis metrics are being applied, CLDs are not typically thought of as analysable as networks, with many papers that apply www.nature.com/scientificreports/ network analysis metrics to CLDs first providing some type of justification for why a CLD could be interpreted as a network (e.g., 8,[23][24][25][27][28][29][30]37,42,43,47,50,54 ).…”

“…In seeking alternative methods, network analysis 35 has gained momentum as an accepted practice to quantitatively identify leverage points based on CLDs 8 , 23 – 30 , 36 – 55 —for example in public health 24 , 38 , 45 , 50 , 52 , 53 , 55 . Here, 22 of the 29 papers that we identified that adopt this practice were published in 2021 (4 papers 23 , 25 , 37 , 50 ), 2022 (8 papers 27 , 28 , 43 – 45 , 47 , 49 , 52 ), and 2023 (10 papers to date 26 , 30 , 36 , 38 , 39 , 41 , 42 , 46 , 53 , 55 ). Network analysis seemingly has the potential to mathematically analyse a CLD, which in this context can be considered a network of factors, to systematically find which factors represent leverage points.…”

“…Currently in the literature relating to the development and analysis of CLDs, the term 'leverage point' tends to be used as referring to a factor within the CLD "in which change is likely to yield significant shifts elsewhere in the system" 23 . The key premise is that the system depicted in the CLD is producing undesirable system-level behaviour, and that there are central drivers in the system that could be influenced in such a way that this undesirable system-level behaviour is sustainably disrupted 8,[24][25][26][27][28][29][30] .…”

“…In seeking alternative methods, network analysis 35 has gained momentum as an accepted practice to quantitatively identify leverage points based on CLDs 8,[23][24][25][26][27][28][29][30][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55] -for example in public health 24,38,45,50,52,53,55 . Here, 22 of the 29 papers that we identified that adopt this practice were published in 2021 (4 papers 23,25,37,50 ), 2022 (8 papers 27,28,[43][44][45]47,49,52 ), and 2023 (10 papers to date …”