As part of my work for a Dutch regional government, I was asked to compare the open data offerings of the 12 provinces. I wanted to use something that levels the playing field for all parties compared and prevents me comparing apples to oranges, so opted for the Dutch national data portal as a source of data. An additional benefit of this is that the Dutch national portal (a CKAN instance) has a well defined API, and uses standardised vocabularies for the different government entities and functions of government.

I am interested in openness, findability, completeness, re-usability, and timeliness. For each of those I tried to pick something available through the API, that can be a proxy for one or more of those factors.

The following aspects seemed most useful:

  • openness: use of open licenses
  • findability: are datasets categorised consistently and accurately so they can be found through the policy domains they pertain to
  • completeness: does a province publish across the entire spectrum of a) national government’s list of policy domains, and b) across all 7 core tasks as listed by the association of provincial governments
  • completeness: does a province publish more than just geographic data (most of their tasks are geo-related, but definitely not all)
  • re-usability: in which formats do provinces publish, and are these a) open standards, b) machine readable, c) structured data

I could not establish a useful proxy for timeliness, as all the timestamps available through the API of the national data portal actually represent processes (when the last automatic update ran), and contain breaks (the platform was updated late last year, and all timestamps were from after that update).

Provinces publish data in three ways, and the API of the national portal makes the source of a dataset visible:

  1. they publish geographic data to the Dutch national geographic register (NGR), from which metadata is harvested into the Dutch open data portal. It used to be that only openly licensed data was harvested but since November last year also closed licensed data is being harvested into the national portal. It seems this is done by design, but this major shift has not been communicated at all.
  2. they publish non-geographic data to dataplatform.nl, a CKAN platform provided as a commercial service to government entities to host open data (as the national portal only registers metadata, and isn’t storing data). Metadata is automatically harvested into the national portal.
  3. they upload metadata directly to the national portal by hand, pointing to specific data sources online elsewhere (e.g. the API of an image library)

Most provinces only publish through the National Geo Register (NGR). Last summer I blogged about that in more detail, and nothing has changed really since then.

I measured the mentioned aspects as follows:

  • openness: a straight count of openly licensed data sets. It is national policy to use public domain, CC0 or CC-BY, and this is reflected in what provinces do. So no need to distinguish between open licenses, just between open and not-openly licensed material
  • findability: it is mandatory to categorise datasets, but voluntary to add more than one category, with a maximum of 3. I looked at the average number of categories per dataset for each province. One only categorises with one term, some consistently provide more complete categorisation, where most end up in between those two.
  • completeness: looking at those same categories, a total of 22 different ones were used. I also looked at how many of those 22 each province uses. As all their tasks are similar, the extend to which they cover all used categories is a measure for how well they publish across their spectrum of tasks. Additionally provinces have self-defined 7 core tasks, to which those categories can be mapped. So I also looked at how many of those 7 covered. There are big differences in the breadth of scope of what provinces publish.
  • completeness: while some 80% of all provincial data is geo-data and 20% non-geographic, less than 1% of open data is non-geographic data. Looking at which provinces publish non-geographic data, I used the source of it (i.e. not from the NGR), and did a quick manual check on the nature of what was published (as it was just 22 data sets out of over 3000, this was still easily done by hand).
  • re-usability: for all provinces I polled the formats in which data sets are published. Data sets can be published in multiple formats. All used formats I judged on being a) open standards, b) machine readable, c) structured data. Formats that matched all 3 got 3 points, that matched machine readable and structure but not open standards 1 points, and didn’t match structure or machine readability no points. I then divided the number of points by the total number of data formats they used. This way you get a score of at most 3, and the closer you get to 3, the more of your data matches the open definition.

As all this is based on the national portal’s API, getting the data and calculating scores can be automated as an ongoing measurement to build a time series of e.g. monthly checks to track development. My process only contained one manual action (concerning non-geo data), but it could be done automatically followed up at most with a quick manual inspection.

In terms of results (which now have been first communicated to our client), what becomes visible is that some provinces score high on a single measure, and it is easy to spot who has (automated) processes in place for one or more of the aspects looked at. Also interesting is that the overall best scoring province is not the best scoring on any of the aspects but high enough on all to have the highest average. It’s also a province that spent quite a lot of work on all steps (internally and publication) of the chain that leads to open data.

Granularity - legos, crayons, and moreGranularity (photo by Emily, license: CC-BY-NC)

A client, after their previous goal of increasing the volume of open data provided, is now looking to improve data quality. One element in this is increasing the level of detail of the already published data. They asked for input on how one can approach and define granularity. I formulated some thoughts for them as input, which I am now posting here as well.

Data granularity in general is the level of detail a data set provides. This granularity can be thought of in two dimensions:
a) whether a combination of data elements in the set is presented in one field or split out into multiple fields: atomisation
b) the relative level of detail the data in a set represents: resolution

On Atomisation
Improving this type of granularity can be done by looking at the structure of a data set itself. Are there fields within a data set that can be reliably separated into two or more fields? Common examples are separating first and last names, zipcodes and cities, streets and house numbers, organisations and departments, or keyword collections (tags, themes) into single keywords. This allows for more sophisticated queries on the data, as well as more ways it can potentially be related to or combined with other data sets.

For currently published data sets improving this type of granularity can be done by looking at the existing data structure directly, or by asking the provider of the data set if they have combined any fields into a single field when they created the dataset for publication.

This type of granularity increase changes the structure of the data but not the data itself. It improves the usability of the data, without improving the use value of the data. The data in terms of information content stays the same, but does become easier to work with.

On Resolution
Resolution can have multiple components such as: frequency of renewal, time frames represented, geographic resolution, or splitting categories into sub-categories or multilevel taxonomies. An example is how one can publish average daily temperature in a region. Let’s assume it is currently published monthly with one single value per day. Resolution of such a single value can be increased in multiple ways: publishing the average daily temperature daily, not monthly. Split up the average daily temperature for the region, into average daily temperature per sensor in that region (geographic resolution). Split up the average single sensor reading into hourly actual readings, or even more frequent. The highest resolution would be publishing real-time individual sensor readings continuously.

Improving resolution can only be done in collaboration with the holder of the actual source of the data. What level of improvement can be attained is determined by:

  1. The level of granularity and frequency at which the data is currently collected by the data holder
  2. The level of granularity or aggregation at which the data is used by the data holder for their public tasks
  3. The level of granularity or aggregation at which the data meets professional standards.

Item 1 provides an absolute limit to what can be done: what isn’t collected cannot be published. Usually however data is not used internally in the exact form it was collected either. In terms of access to information the practical limit to what can be published is usually the way that data is available internally for the data holder’s public tasks. Internal systems and IT choices are shaped accordingly usually. Generally data holders can reliably provide data at the level of Item 2, because that is what they work with themselves.

However, there are reasons why data sometimes cannot be publicly provided the same way it is available to the data holder internally. These can be reasons of privacy or common professional standards. For instance energy companies have data on energy usage per household, but in the Netherlands such data is aggregated to groups of at least 10 households before publication because of privacy concerns. National statistics agencies comply with international standards concerning how data is published for external use. Census data for instance will never be published in the way it was collected, but only at various levels of aggregation.

Discussions on the desired level of resolution need to be in collaboration with potential re-users of the data, not just the data holders. At what point does data become useful for different or novel types of usage? When is it meeting needs adequately?

Together with data holders and potential data re-users the balance needs to be struck between re-use value and considerations of e.g. privacy and professional standards.

This type of granularity increase changes the content of the data. It improves the usage value of the data as it allows new types of queries on the data, and enables more nuanced contextualisation in combination with other datasets.

This article is a good description of the Freedom of Information (#foia #opengov #opendata) situation in the Balkans. Due to my work in the region, I recognise lots of what is described here. My work in the region, such as in Serbia, has let me encounter various institutions willing to use evasive action to prevent the release of information.

In essence this is not all that different from what (decentral) government entities in other European countries do as well. Many of them still see increased transparency and access as a distraction absorbing work and time they’d rather spend elsewhere. Yet, there’s a qualitative difference in the level of obstruction. It’s the difference between acknowledging there is a duty to be transparant but being hesitant, and not believing that there’s such a duty in governance at all.

Secrecy, sometimes in combination with corruption, has a long and deep history. In Central Asia for instance I encountered an example that the number of agricultural machines wasn’t released, as a 1950’s Soviet law still on the books marked it as a state secret (because tractors could be mobilised in case of war). More disturbingly such state secrecy laws are abused to tackle political opponents in Central Asia as well. When a government official releases information based on a transparency regulation, or as part of policy implementation, political opponents might denounce them for giving away state secrets and take them to court risking jail time even.

There is a strong effort to increase transparency visible in the Balkan region as well. Both inside government, as well as in civil society. Excellent examples exist. But it’s an ongoing struggle between those seeing power as its own purpose and those seeking high quality governance. We’ll see steps forward, backwards, rear guard skirmishes and a mixed bag of results for a long time. Especially there where there are high levels of distrust amongst the wider population, not just towards government but towards each other.

One such excellent example is the work of the Serbian information commissioner Sabic. Clearly seeing his role as an ombudsman for the general population, he and his office led by example during the open data work I contributed to in the past years. By publishing statistics on information requests, complaints and answer times, and by publishing a full list of all Serbian institutions that fall under the remit of the Commission for Information of Public Importance and Personal Data Protection. This last thing is key, as some institutions will simply stall requests by stating transparency rules do not apply to them. Mr. Sabic’s term ended at the end of last year. A replacement for his position hasn’t been announced yet, which is both a testament to Mr Sabic’s independent role as information commissioner, and to the risk of less transparency inclined forces trying to get a much less independent successor.

Bookmarked Right to Know: A Beginner’s Guide to State Secrecy / Balkan Insight by Dusica Pavlovic (Balkan Insight)

Governments in the Balkans are chipping away at transparency laws to make it harder for journalists and activists to hold power to account.

There were several points made in the conversation after my presentation yesterday at Open Belgium 2019. This is a brief overview to capture them here.

1) One remark was about the balance between privacy and openness, and asking about (negative) privacy impacts.

The framework assumes government as the party being interested in measurement (given that that was the assignment for which it was created). Government held open data is by default not personal data as re-use rules are based on access regimes which in turn all exclude personal data (with a few separately regulated exceptions). What I took away from the remark is that, as we know new privacy and other ethical issues may arise from working with data combinations, it might be of interest if we can formulate indicators that try to track negative outcomes or spot unintended consequences, in the same way as we are trying to track positive signals.

2) One question was about if I had included all economic modelling work in academia etc.

I didn’t. This isn’t academic research either. It seeks to apply lessons already learned. What was included were existing documented cases, studies and research papers looking at various aspects of open data impact. Some of those are academic publications, some aren’t. What I took from those studies is two things: what exactly did they look at (and what did they find), and how did they assess a specific impact? The ‘what’ was used as potential indicator, the ‘how’ as the method. It is of interest to keep tracking new research as it gets published, to augment the framework.

3) Is this academic research?

No, its primary aim is as a practical instrument for data holders as well as national open data policy makers. It’s is not meant to establish scientific truth, and completely quantify impact once and for all. It’s meant to establish if there are signs the right steps are taken, and if that results in visible impact. The aim, and this connects to the previous question as well, is to avoid extensive modelling techniques, and favor indicators we know work, where the methods are straightforward. This to ensure that government data holders are capable to do these measurements themselves, and use it actively as an instrument.

4) Does it include citizen science (open data) efforts?

This is an interesting one (asked by Lukas of Luftdaten.info). The framework currently does include in a way the existence and emergence of citizen science projects, as that would come up in any stakeholder mapping attempts and in any emerging ecosystem tracking, and as examples of using government open data (as context and background for citizen science measurements). But the framework doesn’t look at the impact of such efforts, not in terms of socio-economic impact and not in terms of government being a potential user of citizen science data. Again the framework is to make visible the impact of government opening up data. But I think it’s not very difficult to adapt the framework to track citizen science project’s impact. Adding citizen science projects in a more direct way, as indicators for the framework itself is harder I think, as it needs more clarification of how it ties into the impact of open government data.

5) Is this based only on papers, or also on approaching groups, and people ‘feeling’ the impact?

This was connected to the citizen science bit. Yes, the framework is based on existing documented material only. And although a range of those base themselves on interviewing or surveying various stakeholders, that is not a default or deliberate part of how the framework was created. I do however recognise the value of for instance participatory narrative inquiry that makes the real experiences of people visible, and the patterns across those experiences. Including that sort of measurements would be useful especially on the social and societal impacts of open data. But currently none of the studies that were re-used in the framework took that approach. It does make me think about how one could set-up something like that to monitor impact e.g. of local government open data initiatives.

Today I gave a brief presentation of the framework for measuring open data impact I created for UNDP Serbia last year, at the Open Belgium 2019 Conference.

The framework is meant to be relatable and usable for individual organisations by themselves, and based on how existing cases, papers and research in the past have tried to establish such impact.

Here are the slides.

This is the full transcript of my presentation:

Last Friday, when Pieter Colpaert tweeted the talks he intended to visit (Hi Pieter!), he said two things. First he said after the coffee it starts to get difficult, and that’s true. Measuring impact is a difficult topic. And he asked about measuring impact: How can you possibly do that? He’s right to be cautious.

Because our everyday perception of impact and how to detect it is often too simplistic. Where’s the next Google the EC asked years ago. but it’s the wrong question. We will only know in 20 years when it is the new tech giant. But today it is likely a small start-up of four people with laptops and one idea, in Lithuania or Bulgaria somewhere, and we are by definition not be able to recognize it, framed this way. Asking for the killer app for open data is a similarly wrong question.

When it comes to impact, we seem to want one straightforward big thing. Hundreds of billions of euro impact in the EU as a whole, made up of a handful of wildly successful things. But what does that actually mean for you, a local government? And while you’re looking for that big impact you are missing all the smaller craters in this same picture, and also the bigger ones if they don’t translate easily into money.

Over the years however, there have been a range of studies, cases and research papers documenting specific impacts and effects. Me and my colleagues started collecting those a long time ago. And I used them to help contextualise potential impacts. First for the Flemish government, and last year for the Serbian government. To show what observed impact in for instance a Spanish sector would mean in the corresponding Belgian context. How a global prediction correlates to the Serbian economy and government strategies.

The UNDP in Serbia, asked me to extend that with a proposal for indicators to measure impact as they move forward with new open data action plans in follow up of the national readiness assessment I did for them earlier. I took the existing studies and looked at what they had tried to measure, what the common patterns are, and what they had looked at precisely. I turned that into a framework for impact measurement.

In the following minutes I will address three things. First what makes measuring impact so hard. Second what the common patterns are across existing research. Third how, avoiding the pitfalls, and using the commonalities we can build a framework, that then in itself is an indicator.Let’s first talk about the things that make measuring impact hard.

Judging by the available studies and cases there are several issues that make any easy answers to the question of open data impact impossible.There are a range of reasons measurement is hard. I’ll highlight a few.
Number 3, context is key. If you don’t know what you’re looking at, or why, no measurement makes much sense. And you can only know that in specific contexts. But specifying contexts takes effort. It asks the question: Where do you WANT impact.

Another issue is showing the impact of many small increments. Like how every Dutch person looks at this most used open data app every morning, the rain radar. How often has it changed a decision from taking the car to taking a bike? What does it mean in terms of congestion reduction, or emission reduction? Can you meaningfully quantify that at all?

Also important is who is asking for measurement. In one of my first jobs, my employer didn’t have email for all yet, so I asked for it. In response the MD asked me to put together the business case for email. This is a classic response when you don’t want to change anything. Often asking for measurement is meant to block change. Because they know you cannot predict the future. Motives shape measurements. The contextualisation of impact elsewhere to Flanders and Serbia in part took place because of this. Use existing answers against such a tactic.

Maturity and completeness of both the provision side, government, as well as the demand side, re-users, determine in equal measures what is possible at all, in terms of open data impact. If there is no mature provision side, in the end nothing will happen. If provision is perfect but demand side isn’t mature, it still doesn’t matter. Impact demands similar levels of maturity on both sides. It demands acknowledging interdependencies. And where that maturity is lacking, tracking impact means looking at different sets of indicators.

Measurements often motivate people to game the system. Especially single measurements. When number of datasets was still a metric for national portals the French opened with over 350k datasets. But really it was just a few dozen, which they had split according to departments and municipalities. So a balance is needed, with multiple indicators that point in different directions.

Open data, especially open core government registers, can be seen as infrastructure. But we actually don’t know how infrastructure creates impact. We know that building roads usually has a certain impact (investment correlates to a certain % rise in GDP), but we don’t know how it does so. Seeing open data as infrastructure is a logical approach (the consensus seems that the potential impact is about 2% of GDP), but it doesn’t help us much to measure impact or see how it creates that.

Network effects exist, but they are very costly to track. First order, second order, third order, higher order effects. We’re doing case studies for ESA on how satellite data gets used. We can establish network effects for instance how ice breakers in the Botnian gulf use satellite data in ways that ultimately reduce super market prices, but doing 24 such cases is a multi year effort.

E puor si muove! Galileo said Yet still it moves. The same is true for open data. Most measurements are proxies. They show something moving, without necessarily showing the thing that is doing the moving. Open data often is a silent actor, or a long range one. Yet still it moves.

Yet still it moves. And if we look at the patterns of established studies, that is what we indeed see. There are communalities in what movement we see. In the list on the slide the last point, that open data is a policy instrument is key. We know publishing data enables other stakeholders to act. When you do that on purpose you turn open data into a policy instrument. The cheapest one you have next to regulation and financing.

We all know the story of the drunk that lost his keys. He was searching under the light of a street lamp. Someone who helped him else asked if he lost the keys there. No, the drunk said, but at least there is light here. The same is true for open data. If you know what you published it for, at least you will be able to recognise relevant impact, if not all the impact it creates. Using it as policy instrument is like switching on the lights.

Dealing with lack of maturity means having different indicators for every step of the way. Not just seeing if impact occurs, but also if the right things are being done to make impact possible: Lead and lag indicators

The framework then is built from what has been used to establish impact in the past, and what we see in our projects as useful approaches. The point here is that we are not overly simplifying measurement, but adapt it to whatever is the context of a data provider or user. Also there’s never just one measurement, so a balanced approach is possible. You can’t game the system. It covers various levels of maturity from your first open dataset all the way to network effects. And you see that indicators that by themselves are too simple, still can be used.

Additionally the framework itself is a large scale sensor. If one indicator moves, you should see movement in other indicators over time as well. If you throw a stone in the pond, you should see ripples propagate. This means that if you start with data provision indicators only, you should see other measurements in other phases pick up. This allows you to both use a set of indicators across all phases, as well as move to more progressive ones when you outgrow the initial ones.finally some recommendations.

Some final thoughts. If you publish by default as integral part of processes, measuring impact, or building a business case is not needed as such. But measurement is very helpful in the transition to that end game. Core data and core policy elements, and their stakeholders are key. Measurement needs to be designed up front. Using open data as policy instrument lets you define the impact you are looking for at the least. The framework is the measurement: Only micro-economic studies really establish specific economic impact, but they only work in mature situations and cost a lot of effort, so you need to know when you are ready for them. But measurement can start wherever you are, with indicators that reflect the overall open data maturity level you are at, while looking both back and forwards. And because measurement can be done, as a data holder you should be doing it.

US Congress just before leaving for Christmas has voted to approve a new law, that mandates two key elements: public information is open by default and needs to be made actively available in machine readable format, as well as that policy making should be evidence based. In order for agencies to comply they will need to appoint a Chief Data Officer.

I think while of those two the first one (open data) is the more immediately visible, the second one, about evidence based policy making, is much more significant long term. Government, especially politics, often is willingly disinterested in policy impact evaluation. It’s much more status enhancing to announce new plans than admitting previous plans didn’t come to anything. Evidence based policy will help save money. Additionally government agencies will soon realise that doing evidence based policy making is made a lot easier if you already do open data well. The evidence you need is in that open data, and it being open allows all of us to go look for that evidence or its absence.

There’s one caveat to evidence based policy making: it runs the risk of killing any will to experiment. After all, by definition there’s no evidence for something new. So a way is needed in which new policies can be tried out as probes. To see if there’s emerging evidence of impact. Again, that evidence should become visible in existing open data streams. If evidence is found the experimental policy can be rolled out more widely. Evidence based policies need experiments to help create an evidence base, not just of what works but also of what doesn’t.

A great result for the USA’s open government activists. This basically codifies the initiatives of the Obama Presidency, which were the trigger for much of the global open data effort these last 10 years, into US federal law.