" A good diagnosis points to a treatment. The Commission has measured something other than the disease and prescribed more visits. "
Funny, so topical. The irony of that specific metaphor, is that NZ Parliament parties appear to be effectively prescribing more visits, but doing very little to address supply, creating demand for a service which doesn't exist with no insights let alone plans on how to help. By focusing on monitoring rather than eliminating the regulatory barriers holding back new entrants, they’ve trapped Kiwi markets in a loop: more bureaucracy to treat the side effects of bureaucracy.
- The actual concern from parents/clinicians are to be able to access assessment and supports (e.g. schooling, respite care, etc.).
- The concern for autistic adults is typically discrimination / social inclusion.
But, Governments have heard, "let's help diagnose autism". And there's lots of people with an opinion, including many well-meaning clinicians who offer premium autism assessment services in private and may have a little cognitive dissonance around statement and management of interest. Policymakers are given ye old, "more with less", and try to figure out how to have high impact at low cost, often without understanding the domain let alone the specific issue, let alone the economic mechanisms available to them to influence. So, they've almost universally said (with one very notable exception, Peter Malinauskas, who is a very smart man dressed up as a street hustler), "why don't we screen for it more".
So policymakers fund more screening. It's typically in schools and early childhood (services like Plunket). And so more people are suspected to have autism, more people screen positive, the index of suspicion rises, and further converts latent into realised demand. Screen positive people are referred to health services, who have received no investment to handle increased referrrals. Funny thing is, a number of these policymakers are noted to have said things like, "The problem with healthcare is health managers", ignoring that they were often warned of the consequences but thought they knew better.
Peeps are referred to disability services who say, "we can't treat them until they're diagnosed". Then hobbiest policymakers come along and say, "let's help diagnose autism", and decide to introduce more screening... And so it goes.
Health is simply the language of my response, but it's analogous because whether one sits on the right and uses the language of business, or one sits on the left and uses the language of national provision of human services, these are complex systems where the problems typically have clear albeit dynamic causal models for which experts exist.
LLMs may well be a very useful policy governance tool to prevent the boom-and-bust of "truthy" policy design and political, but not understanding the causal mechanisms for complex, impactful and very expensive problems, means that more screening and monitoring will simply tell you more of what you already know, not how to address the issues.
So they have proven themselves to be incompetent and incapable of hypothesising a structural calculated proposition to assist in planning any useful positive outcomes.
Just keeping their seats warm and their grossly unjustifiable unearned inequitable pensions ever closer within reach.
Same old 'public service' by name, highway robbery by nature.
Another excellent post thanks Roger. I particularly enjoyed this extract:
". Competition economists have understood for more than forty years that the number of firms in a market is a weaker signal of competitive intensity than the conditions of entry. A market with three firms and no barriers can be more competitive than a market with eight firms and high barriers. Incumbents discipline themselves against the threat of entry, not only against the firms already present. The report measures the firms. It does not measure the barriers."
Another point you (and some the comments to your post) make is the old adage: "There are lies, damn lies and statistics".
" The categories tell you nothing, because beverages include both water and vodka, and snacks include both carrot sticks and deep-fried Mars bars. "
I have a couple of series, not just posts, on nosology-the principle ontological and epistemic mechanism of medicine. It's become a hot topic, particularly with Awais Aftab's (https://substack.com/@awaisaftab) recent article in the New York Times (https://www.nytimes.com/2026/05/11/opinion/adhd-autism-depression-diagnoses.html) which in some ways is the canonical article of our times, explaining and describing the impacts of medical reification.
Also known as hypostatization or the fallacy of misplaced concreteness, it is the logical error of treating an idea (like "justice" or "society") as a tangible entity. I can't help feel that by definition this is worse in legal positivism, where law defines rules, and so almost their entire social purpose is to define discrete boundaries.
Medicine, like computer science, loves ontologies. In fact, combine them and you create bioinformatics, where you basically have to create an ontology to get your PhD (https://academic.oup.com/bioinformatics). My favourite is of course the Rare Diseases Ontology (https://sciences.orphadata.com/ordo/).
I hadn't really considered food ontologically. I think what you're describing is that CC appear to have uncritically used administrative data, not recategorizing it or normalising to an ontology. I'm not certain about economics, but this is sort of Epidemiology 101 (think of epidemiology as the "data science" of diseases). NZ has many excellent epidemiologists who are expert in this area that they probably could have just asked.
As a simple example, in cost-of-illness work, we analyze and report descriptive, comparative and timeseries descriptors for medicine classes. But they vary by country, so in parallel I use the WHO's ATC ontology (https://www.who.int/tools/atc-ddd-toolkit/atc-classification), which is a really excellent example because it has both high and low-level abstractions.
The methods used ANZSIC (Australian and New Zealand Standard Industrial Classification) and ANCO (ANCO (Australian and New Zealand Commodity Classification), presumably because those are standard for StatsNZ and IDI.
The report specifically notes that aggregate macro-metrics can blunt or obscure individual market realities, and it explicitly highlights a lack of "market-specific" detail. If the data were to be deeply normalised specifically to trace food and grocery economics, these frameworks would be required to bridge the gap between Stats NZ's business data and microeconomic retail realities, thereby at least providing a comparative basis.
National-level analyses that are not normalised to multiple analogies, often become effectively meaningless I'd need to check with Eric or others, but I'm pretty sure this is one of the main reasons that organisations like the OECD exist, to facilitate these comparative insights.
Because of CC's work, I'd have suggested:
- GS1 Global Product Classification (GPC): given their active work in monitoring and providing decision-analytic modelling on the grocery sector.
- Central Product Classification (CPC): The UN-maintained CPC maps goods directly into national accounts. Because CC's report heavily weighs the impact of global trade and imports on domestic competition, mapping the ANZSIC data to the semantic version of CPC would allow them to calculate exactly how domestic food production values compete with international price indexes.
- Harmonized System (HS) Codes: Tracing global food trade expenditures and the impact of tariff margins on Kiwis.
- Valueflows Ontology: As a semantic web tool designed for tracking economic networks, it could be used to connect Stats NZ’s "upstream/downstream" business dynamism data with actual food distribution paths- quantifying the exact cost accumulation as food moves from New Zealand farms to processing plants->retail duopolies->consumer.
In the age of generative and agentic AI, I think researchers could be excused for not doing doing all of this. But I don't think the peak commerce governance organisation can really excuse it- to be clear, I'm not blaming anyone- I'd bet their advisers told decision-makers. It's a lot of resources to have spent on a project without having produced these insights.
Putting aside that a quick search of one's LLM of the day could have helped with this, if implementing those methods in the IDI were a concern, there are three further considerations:
- Firstly, I'm not entirely certain why the StatsNZ/IDI maintains such a restrictive approach to tool usage. Being able to use a broader range of tools would make increase the value of the investment at little to no additional cost.
- Secondly, the persistent belief that airlocked on-prem/private cloud is the way to provide the IDI, is short-sighted. It's one of the areas where industry is really orders of magnitudes ahead of most governments (except perhaps for signals intelligence). Platforms like MS Purview are revolutionary for linked administrative analysis, and their most government agencies simply couldn't even come close in security to large public cloud providers, combined which manage the largest amount of classified data on the planet.
- Thirdly, any concern about the use of AI or coding agents in the IDI is easily addressed- use a local open weights model like LFM2.5 (https://arxiv.org/html/2511.23404v1), doesn't require many resources and can be managed entirely airlocked.
well done, as always, thanks Roger
" A good diagnosis points to a treatment. The Commission has measured something other than the disease and prescribed more visits. "
Funny, so topical. The irony of that specific metaphor, is that NZ Parliament parties appear to be effectively prescribing more visits, but doing very little to address supply, creating demand for a service which doesn't exist with no insights let alone plans on how to help. By focusing on monitoring rather than eliminating the regulatory barriers holding back new entrants, they’ve trapped Kiwi markets in a loop: more bureaucracy to treat the side effects of bureaucracy.
There is a specific and pointed example in both Australia and New Zealand healthcare: people are concerned about autism (https://rareinsights.substack.com/p/can-we-ever-know-how-many-autistic):
- The actual concern from parents/clinicians are to be able to access assessment and supports (e.g. schooling, respite care, etc.).
- The concern for autistic adults is typically discrimination / social inclusion.
But, Governments have heard, "let's help diagnose autism". And there's lots of people with an opinion, including many well-meaning clinicians who offer premium autism assessment services in private and may have a little cognitive dissonance around statement and management of interest. Policymakers are given ye old, "more with less", and try to figure out how to have high impact at low cost, often without understanding the domain let alone the specific issue, let alone the economic mechanisms available to them to influence. So, they've almost universally said (with one very notable exception, Peter Malinauskas, who is a very smart man dressed up as a street hustler), "why don't we screen for it more".
So policymakers fund more screening. It's typically in schools and early childhood (services like Plunket). And so more people are suspected to have autism, more people screen positive, the index of suspicion rises, and further converts latent into realised demand. Screen positive people are referred to health services, who have received no investment to handle increased referrrals. Funny thing is, a number of these policymakers are noted to have said things like, "The problem with healthcare is health managers", ignoring that they were often warned of the consequences but thought they knew better.
Peeps are referred to disability services who say, "we can't treat them until they're diagnosed". Then hobbiest policymakers come along and say, "let's help diagnose autism", and decide to introduce more screening... And so it goes.
Health is simply the language of my response, but it's analogous because whether one sits on the right and uses the language of business, or one sits on the left and uses the language of national provision of human services, these are complex systems where the problems typically have clear albeit dynamic causal models for which experts exist.
LLMs may well be a very useful policy governance tool to prevent the boom-and-bust of "truthy" policy design and political, but not understanding the causal mechanisms for complex, impactful and very expensive problems, means that more screening and monitoring will simply tell you more of what you already know, not how to address the issues.
So they have proven themselves to be incompetent and incapable of hypothesising a structural calculated proposition to assist in planning any useful positive outcomes.
Just keeping their seats warm and their grossly unjustifiable unearned inequitable pensions ever closer within reach.
Same old 'public service' by name, highway robbery by nature.
Another excellent post thanks Roger. I particularly enjoyed this extract:
". Competition economists have understood for more than forty years that the number of firms in a market is a weaker signal of competitive intensity than the conditions of entry. A market with three firms and no barriers can be more competitive than a market with eight firms and high barriers. Incumbents discipline themselves against the threat of entry, not only against the firms already present. The report measures the firms. It does not measure the barriers."
Another point you (and some the comments to your post) make is the old adage: "There are lies, damn lies and statistics".
" The categories tell you nothing, because beverages include both water and vodka, and snacks include both carrot sticks and deep-fried Mars bars. "
I have a couple of series, not just posts, on nosology-the principle ontological and epistemic mechanism of medicine. It's become a hot topic, particularly with Awais Aftab's (https://substack.com/@awaisaftab) recent article in the New York Times (https://www.nytimes.com/2026/05/11/opinion/adhd-autism-depression-diagnoses.html) which in some ways is the canonical article of our times, explaining and describing the impacts of medical reification.
Also known as hypostatization or the fallacy of misplaced concreteness, it is the logical error of treating an idea (like "justice" or "society") as a tangible entity. I can't help feel that by definition this is worse in legal positivism, where law defines rules, and so almost their entire social purpose is to define discrete boundaries.
The risk of course is that the define boundaries that are imaged rather than real, and without trying to derail your article, this is the biggest epistemic and logical issue with this proposed law: https://www3.parliament.nz/en/pb/sc/make-a-submission/document/54SCSSC_SCF_9E8E8A14-A51C-4567-AB33-08DE9053A7D1/legislation-definitions-of-woman-and-man-amendment-bill.
Medicine, like computer science, loves ontologies. In fact, combine them and you create bioinformatics, where you basically have to create an ontology to get your PhD (https://academic.oup.com/bioinformatics). My favourite is of course the Rare Diseases Ontology (https://sciences.orphadata.com/ordo/).
I hadn't really considered food ontologically. I think what you're describing is that CC appear to have uncritically used administrative data, not recategorizing it or normalising to an ontology. I'm not certain about economics, but this is sort of Epidemiology 101 (think of epidemiology as the "data science" of diseases). NZ has many excellent epidemiologists who are expert in this area that they probably could have just asked.
As a simple example, in cost-of-illness work, we analyze and report descriptive, comparative and timeseries descriptors for medicine classes. But they vary by country, so in parallel I use the WHO's ATC ontology (https://www.who.int/tools/atc-ddd-toolkit/atc-classification), which is a really excellent example because it has both high and low-level abstractions.
The methods used ANZSIC (Australian and New Zealand Standard Industrial Classification) and ANCO (ANCO (Australian and New Zealand Commodity Classification), presumably because those are standard for StatsNZ and IDI.
The report specifically notes that aggregate macro-metrics can blunt or obscure individual market realities, and it explicitly highlights a lack of "market-specific" detail. If the data were to be deeply normalised specifically to trace food and grocery economics, these frameworks would be required to bridge the gap between Stats NZ's business data and microeconomic retail realities, thereby at least providing a comparative basis.
National-level analyses that are not normalised to multiple analogies, often become effectively meaningless I'd need to check with Eric or others, but I'm pretty sure this is one of the main reasons that organisations like the OECD exist, to facilitate these comparative insights.
Because of CC's work, I'd have suggested:
- GS1 Global Product Classification (GPC): given their active work in monitoring and providing decision-analytic modelling on the grocery sector.
- Central Product Classification (CPC): The UN-maintained CPC maps goods directly into national accounts. Because CC's report heavily weighs the impact of global trade and imports on domestic competition, mapping the ANZSIC data to the semantic version of CPC would allow them to calculate exactly how domestic food production values compete with international price indexes.
- Harmonized System (HS) Codes: Tracing global food trade expenditures and the impact of tariff margins on Kiwis.
- Valueflows Ontology: As a semantic web tool designed for tracking economic networks, it could be used to connect Stats NZ’s "upstream/downstream" business dynamism data with actual food distribution paths- quantifying the exact cost accumulation as food moves from New Zealand farms to processing plants->retail duopolies->consumer.
In the age of generative and agentic AI, I think researchers could be excused for not doing doing all of this. But I don't think the peak commerce governance organisation can really excuse it- to be clear, I'm not blaming anyone- I'd bet their advisers told decision-makers. It's a lot of resources to have spent on a project without having produced these insights.
Putting aside that a quick search of one's LLM of the day could have helped with this, if implementing those methods in the IDI were a concern, there are three further considerations:
- Firstly, I'm not entirely certain why the StatsNZ/IDI maintains such a restrictive approach to tool usage. Being able to use a broader range of tools would make increase the value of the investment at little to no additional cost.
- Secondly, the persistent belief that airlocked on-prem/private cloud is the way to provide the IDI, is short-sighted. It's one of the areas where industry is really orders of magnitudes ahead of most governments (except perhaps for signals intelligence). Platforms like MS Purview are revolutionary for linked administrative analysis, and their most government agencies simply couldn't even come close in security to large public cloud providers, combined which manage the largest amount of classified data on the planet.
- Thirdly, any concern about the use of AI or coding agents in the IDI is easily addressed- use a local open weights model like LFM2.5 (https://arxiv.org/html/2511.23404v1), doesn't require many resources and can be managed entirely airlocked.