Ofqual v Information Commissioner

This appeal is about ‘the algorithm’. Every teenager who was supposed to sit GCSEs and A levels in 2020 and 2021, their parents, schools, universities, potential employers and anyone who took even a passing interest in the news during the 2020 lockdown and the months following will surely remember what that meant. But time passes and memories fade, so a brief explanation may be in order.

The First-tier Tribunal set out a detailed history in its decision. These are the key events for the purposes of this appeal. On 18 March 2020, the Secretary of State for Education decided that no assessments or examinations would take place that academic year. On 31 March, he directed the Office of Qualifications and Examination Regulation (Ofqual from now on) that: ‘students should be issued with calculated results based on their exam centre’s judgements of their ability in the relevant subjects, supplemented by a range of other evidence’. The process would involve a statistical standardisation using historical data for the distribution of grades in subjects for each individual education centre. This was the algorithm. It operated differently depending on the number of students taking each subject. Its purpose was two-fold. First, it was designed to reflect the distribution of grades in previous years. Second, it was designed to reduce the effect of leniency or severity by individual teachers.

Policy continued to develop and, on 13 August 2020, the Secretary of State directed Ofqual to allow students to appeal on the basis of having achieved a higher grade in their mock examinations. That date was the day that A level results were announced. On 16 August, Ofqual decided that students would be given the higher of the centre assessed grade or the calculated grade. On 17 August, the Secretary of State made an announcement, which contained this passage:

This case starts with the requests made to Ofqual under the Freedom of Information Act 2000 (FOIA from now on). They were made under the name of Hedro Pedro. That person did not take any part in the proceedings before either the First-tier Tribunal or this tribunal.

I say ‘requests’ because the terms of the request were varied or clarified a number of times. The First-tier Tribunal decided that the relevant request was made on 9 October 2020:

The Information Commissioner received a complaint under section 50 FOIA on 12 November 2020. On 5 August 2021, the Commissioner decided that although section 36(2)(c) was engaged, the public interest balance under section 2 required Ofqual to disclose the information. This led to Ofqual’s appeal to the First-tier Tribunal. The tribunal dismissed the appeal, but I gave Ofqual permission to appeal to the Upper Tribunal.

These are the provisions of FOIA that I refer to in this decision.

Application for decision by Commissioner

Any person (in this section referred to as “the complainant”) may apply to the Commissioner for a decision whether, in any specified respect, a request for information made by the complainant to a public authority has been dealt with in accordance with the requirements of Part I.

Appeal against notice served under Part IV

Where a decision notice has been served, the complainant or the public authority may appeal to the Tribunal against the notice.

A public authority on which an information notice or an enforcement notice has been served by the Commissioner may appeal to the Tribunal against the notice.

Determination of appeals

If on an appeal under section 57 the Tribunal considers—

that the notice against which the appeal is brought is not in accordance with the law, or

to the extent that the notice involved an exercise of discretion by the Commissioner, that he ought to have exercised his discretion differently,

On such an appeal, the Tribunal may review any finding of fact on which the notice in question was based.

The tribunal analysed the factors accepted by Ofqual as relevant to the public interests balance. This is a summary. It is not intended to be comprehensive.

Ofqual argued that there was a public interest in promoting the integrity of awarding qualifications through publication of centre-level awarding information. It argued that this should carry limited weight. The tribunal decided that no weight should be given to this factor, because the information requested would not shed any light on the behaviour and integrity of an educational centre’s determination.

Ofqual argued that there was a public interest in disclosing information relevant to the decision-making of a public authority. It argued that this should carry low weight. The tribunal accepted that disclosure would promote transparency and openness. In general terms, those factors would carry limited weight. However, the Information Commissioner submitted that the issue was not generality, but the specific public interest in open and transparent information about the impact of the algorithm. The tribunal quoted from the Secretary of State’s announcement on 17 August. Ofqual argued that the algorithm did operate fairly. Their witness did, though, concede that ‘the system’ had produced ‘too many inconsistent outcomes’ which resulted in a loss of ‘public confidence’. This loss of confidence was what led Ofqual to change its approach. The tribunal concluded that the change in the method of awarding A level grades was of personal interest to a large number of students and teachers, and attracted a significant amount of interest from the public at large both in August 2020 (when the change took place) and November 2020 (when Ofqual responded to the request). The tribunal also decided that a better understanding of the fairness of the algorithm could only have enhanced the debate on the system that would be applied in 2021.

Ofqual argued that disclosing the requested information would add little to the debate that had already taken place. The tribunal pointed out that much of the information that had been discussed had only become public after 5 November 2020. The tribunal limited its consideration to information that was already in the public domain on that date.

Overall, the tribunal’s conclusion was that disclosing the information would add significant value to the debate about the 2020 grading system and inform the public interest in understanding whether the algorithm operated fairly. The public interest in disclosure was very strong.

Ofqual identified four factors in favour of maintaining the exemption.

First, Ofqual argued that disclosure would present a misleading impression and lead to reputational damage. The tribunal accepted that reliable conclusions about the integrity of assessments by individual centres could not be drawn from the information. Irresponsible reporting could be guarded against by a simple warning. Centres could be prejudiced by such reporting, but the public were generally aware of the unique circumstances obtaining in 2020.

Second, Ofqual argued that disclosure would be prejudicial to students who were awarded grades in 2020. As to employers, the tribunal considered that any prejudice against students would be set against the known fact that the 2020 system had an inflationary effect on grades. As to universities, any prejudice would be remote, given their sophisticated methods of selecting candidates.

Third, Ofqual argued that disclosure would be prejudicial to their relationship with stakeholders (the Department for Education and the examination boards). The argument was that disclosure would undermine assurances about confidentiality of centre-level grading and undermine trust in Ofqual and its relationships with its stakeholders. In particular, the Secretary of State had made statements that there would be no publication of performance data. The tribunal accepted that there was in general a significant public interest in a relationship of trust between Ofqual and its stakeholders. This required a real-world assessment, involving how commitments would have been reasonably understood and what commitments, so understood, had been breached. The tribunal did not find it helpful to consider what was and was not performance data. The commitment meant that institutional level data that could be the basis of comparisons between different centres would not be disclosed. The disclosure of the information would not allow that. Disclosure could have a negative impact on relationships but it should not be accorded significant weight.

Fourth, Ofqual argued that disclosure would be prejudicial to the integrity of grading in future years. The tribunal accepted that there was a possibility that some teachers might view the disclosure as a breach of the assurance not to publish performance data and would feel under undue pressure to such an extent that it could affect the accuracy of their contribution to grading for 2021. However, the tribunal assessed the risk of this as low. The tribunal also took into account that any future system would have a quality assurance designed to ensure the integrity of the system.

This was the tribunal’s conclusion:

There were four grounds of appeal, divided into nine issues. I have decided that the appeal succeeds on Issue 6 under Ground 3. I do, though, deal with two Issues that need to be resolved before the rehearing. They are Issue 1 under Ground 1 and Issue 8 under Ground 4. Otherwise, it is not necessary to deal with the other Issues; any errors that were made will be subsumed by the rehearing.

Ground 3 is:

Issue 6 is:

I do not consider that the tribunal’s conclusion was a finding of fact. I would prefer to call it a judgment or assessment and have approached this issue on that basis. Whichever label is correct, the tribunal’s reasons do not explain or justify its conclusion.

The tribunal made this judgment in the course of considering the factors accepted by Ofqual as favouring disclosure of the requested information. It first considered at [74]-[75] whether disclosure would promote ‘the integrity of awarding qualifications through publication of centre-level awarding information.’ Ofqual argued that this should be given ‘limited weight’. The tribunal decided that ‘on the facts of this case no weight can be attached’ to it.

The tribunal went on to consider whether disclosure was ‘relevant to the decision-making of a public authority’. Ofqual argued that the weight of this factor was low. The tribunal decided at [76] that in general terms transparency and openness were of limited weight. However, it also recorded at [76] the Information Commissioner’s argument, which concerned ‘the specific public interest in open and transparent information about the impact of the Algorithm.’ The argument was that the information ‘would add to the public debate both as to the fairness of the Algorithm and as to whether the CAG model was a fairer method of awarding A level grades in 2020 than the model used to obtain the calculated grades.’ That was recorded as the Commissioner’s argument. Ofqual has treated the tribunal as accepting the argument.

The tribunal next referred at [77]-[78] to the Secretary of State’s statement on 17 August 2020 and its criticisms of the algorithm. It noted that Ofqual did not accept that the algorithm had operated unfairly, but did accept that it had to be abandoned as a result of the loss of public confidence.

The tribunal decided at [79] that the method of awarding grades in 2020 was ‘a matter of personal interest to a large number of students and teachers, and also attracted a significant amount of interest from the public at large’ both in August 2020 and at the time of the request in November.

That was the end of the first part of the tribunal’s analysis. It went on to consider at [80] the value of the information to the debate taking place in November 2020 about the ‘mechanism for the award of A level grades in 2021’. It concluded that ‘a better understanding of the fairness of the Algorithm in November 2020 would only have enhanced the consideration and debate as to an appropriate alternative mechanism for awarding A level grades in 2021, the need for which … was a possibility at the relevant date.’

Ofqual argued that the information would add little to the public interest in view of the information in the public domain. The tribunal referred at [84] to and commented on some of the information already in the public domain as of 5 November 2020, but rejected at [85] information that was published after that date. As to the latter, it was right to refuse to take that information into account. See my analysis of Ground 1, Issue 1 below.

The tribunal’s overall conclusion at [86] was that the information ‘would, as of 5 November 2020, have added significant value to any debate regarding the 2020 A level grading system. … [It] would help inform the public interest in understanding whether the Algorithm operated fairly. We find this public interest to be very strong.’

The starting point is the nature and content of the information. It is statistical, so disclosure would only show the scale and range of the changes made as a result of abandoning reliance on the algorithm. It would therefore show the effect of using the algorithm in the first place. It would thereby add content to the Secretary of State statement that the algorithm had produced ‘too many inconsistent and unfair outcomes’ and revealed ‘a number of anomalies that had not been anticipated by Ofqual and which severely undermined confidence in the system.’

How did that further the public interest identified by the tribunal? The tribunal identified the only public interest as the debate on fairness. I assume that means the accuracy of the algorithm’s assessment of a pupil’s ability or likely performance. It is not clear to me whether the tribunal accepted the Information Commissioner’s formulation in [76] or not. That probably does not matter, as that formulation is also about fairness. What the tribunal did not explain is how the statistical information about the scale of the effect of the algorithm provided a very strong public interest in disclosure. If the issue is fairness, the scale of the effect of the algorithm would only be relevant if that effect was unfair or anomalous. But the scale of the effect would not help to show whether it was having those effects.

It is possible that the statistical information had value when considered in combination with other evidence in the public domain at the time. The tribunal did comment at [84] on some of that other evidence. Its comments, though, do not show how the statistical information and that other evidence in combination justified its assessment of the strength of the public interest. If the information had that effect on its own, the reason is not evident to me.

That is sufficient to decide this appeal. The nature and strength of the public interest in disclosure is the comparator against which the public interest in maintaining an exemption must be judged. If there was a mistake in the former, the comparison is inevitably undermined.

Ground 1 is:

Issue 1 is:

The answer to this issue depends on the interpretation and application of section 2(2)(b) FOIA.

Montague v Information Commissioner [2022] UKUT 104 (AAC) [2023] 1 WLR 1565 was a decision of a three-judge panel of the Upper Tribunal. As such, it is binding on me for any point of law it decides. If it is wrong, the mistake will have to be corrected by the Court of Appeal.

Montague dealt with two principle issues. Only the second is relevant to this case:

As to the Public Interest Timing Issue, we conclude it is to be judged at the time the public authority makes its decision on the request which has been made to it and that decision making time does not include any later decision made by the public authority reviewing a refusal decision it has made on the request.

… the answer to when the public interest considerations fall to be judged is to be provided by construing the relevant statutory provisions in context.

… The public authority is not to be judged on the balance of competing public interests on how matters stand other than at the time of the decision on the request which it has been obliged by Part I of FOIA to make.

Mr Kosmin emphasised the opening words of subparagraph (b): ‘in all the circumstances of the case’. I accept that section 12(2)(b) refers to ‘all the circumstances of the case’, but that does not identify the time as at which the test must be applied. The public authority must take account of all the circumstances of the case that are relevant at that time. Montague decided what that time was. I accept that the factual background in that case was different from this. It is important, though, to see how the panel analysed the issue it had to decide. It was argued as an issue of statutory interpretation and that is how it was analysed by the tribunal. Which is why the tribunal referred in paragraph 48 to construing the relevant statutory provisions ‘in context’, a standard reference to the statutory context of section 12(2)(b). The analysis was inevitably carried out in the context of the facts of that case and expressed in terms appropriate to that context. Which is why the tribunal referred to the evidence becoming available ‘well after’ the time when the public authority made its decision. But that does not alter the conclusion or the principle of law for which the case is an authority. Given its analysis and conclusion, it set out the principle that the First-tier Tribunal applied in this case. There is no getting around the clarity of paragraph 86 of the panel’s reasons.

Even if I were free to do so, I would not accept Mr Kosmin’s argument that a public authority can take account of information that is known will enter the public domain imminently. On his argument, this would include both information that would be published by the public authority itself and information that would be published by another public authority. This formulation raises vague questions. What is imminent? How certain must it be that specific information will be published? What happens if, in the event, no information or different information is published? How does the requester know at the time whether the public authority correctly assessed the balance of interests on account of information that is not yet available? Questions like this render the proposed test impracticable in application.

To summarise, this is the position. Section 2(2)(b) has to be applied in all the circumstances of the case as at the date when a public authority decides how to respond to a request for information. On review or appeal, information can be taken into account that was not known to or not available to the public authority at that date, but only in so far as it is relevant to the decision that should have been made on that date. This approach is not novel. It applies, for example, when deciding on an appeal: (a) whether there was ‘a risk of serious personal injury’ under section 22(2) of the Health and Safety at Work etc Act 1974 (HM Inspector of Health and Safety v Chevron North Sea Ltd [2018] 1 WLR 964 at [18]); and (b) whether a circumstance was obtaining at the time of a decision for the purposes of section 12(8)(b) of the Social Security Act 1998 (R(DLA) 2 and 3/01). That is the approach that the First-tier Tribunal took and that it must take at the rehearing.

Ground 4 is:

Issue 8 is:

This ground concerns the evidence from Ofqual and the Secretary of State on the importance of the reputational damage that might occur if the information were disclosed. Mr Kosmin argued that there was information that was available to Ofqual and Secretary of State but not to the tribunal, and that the expertise or experience possessed by Ofqual and the Secretary of State was not possessed by the tribunal.

Mr Kosmin relied on paragraph 56 of the Upper Tribunal’s decision in APPGER (No 1). This is the full paragraph:

These are my conclusions. First, there is no precautionary principle that applies to a particular category of cases or requires a special approach to particular issues. The authorities do not support it. Second, there is no need for it. The conclusions in the cases relied on can be achieved on basic principles of assessing evidence. Third, the existence of such a principle would cause unnecessary problems.

The starting point is, as it always must be, the tribunal’s jurisdiction and powers. Legislation may exclude specified issues from the tribunal’s jurisdiction, as it does in section 4(3) of the Safeguarding Vulnerable Groups Act 2006 by excluding the issue of appropriateness (discussed in Disclosure and Barring Service v JHB [2023] EWCA Civ 982). Or it can limit the scope of that jurisdiction, as it does in section 36(2) FOIA by restricting its scope by reference to the ‘reasonable opinion of a qualified person’. That restriction does not apply in this case because the information was statistical (see section 36(4)), but my point is that this is one of the techniques available to control the scope of a tribunal’s jurisdiction.

In FOIA, the First-tier Tribunal’s jurisdiction is governed by sections 57 and 58. Section 58 governs the issues on which the tribunal may allow an appeal. Both sections lead back to the Information Commissioner’s decision notice under section 50, which in turn leads back to public authority’s duty under section 2(2)(b). There is no statutory restriction on the tribunal’s jurisdiction or powers requiring it to accept the view of any witness or party on any issue. It would be abdicating its judicial authority if it were to do so.

What the tribunal has to do is to assess the evidence and arguments. This is governed by the general law. My reading of the authorities cited is that they are no more than instances of the application of general principles to particular circumstances of those cases. This is how the Supreme Court dealt with the issue in R (Begum) v Special Immigration Appeals Commission [2021] AC 765, which concerned the power of the Secretary of State for the Home Department to exclude people from entry into the United Kingdom.

When a tribunal assesses any evidence, it must take account of any factors that enhance or detract from the quality of the evidence. Those factors include the particular advantages available to the witness that are not available to the tribunal. That factor is most frequently discussed in relation to expert evidence, where almost by definition the witness knows more than the tribunal. Despite that advantage, the tribunal is responsible for finding the facts and may reject the opinion of even the most distinguished expert, although it must have a reason for doing so (Re B (Care: Expert Witness) [1996] 1 FLR 667 at 670 and 674). Ultimately, the value of evidence depends on the reasoning, not the conclusion (Kennedy v Cordia (Services) LLP [2016] 1 WLR 597 at [48]). I accept Ms Taunton’s argument that this issue is ‘in substance, a weight challenge.’

This approach avoids any need to define principles or draw the boundaries within which they apply. It is flexible enough to take account of the potential mix and interaction of different factors that may be present. It recognises the spectrum of cases and issues that can arise. Cases like Begum and APPGER (No 1) are at one end of that spectrum, involving as they do issues of international relations and national security. At the opposite extreme is the advantage a witness in a personal injuries’ action may have had by seeing a collision from a particular vantage point. The general approach to assessing evidence can take account of both extremes and the infinite variety in between. Boundaries and categories are not helpful and introduce unnecessary complications.

That is all I need to say. Any mistake that may have been made in assessing the evidence to which Mr Kosmin referred or in explaining that tribunal’s reasoning will be subsumed by the rehearing.