Setting grade boundaries is very difficult. In reality, even the actual published grade boundaries for a specific exam paper may not really be valid for a different cohort.
Add in the changes to both the content and the grading systems for most GCSEs, and we have a very complex situation.
Some important points
Internal consistency
If the marking is accurate and has been moderated, it should be internally consistent. I.e. the students who gets the most marks has the best attainment.
This means you will always be able to compare the attainment of individuals and groups within the subject.
External pegging
The problem comes when you try to link a set of marks to an actual external real life GCSE (or other) grade.
There are some assumptions we can make that will help
 We know the grade equivalencies. The bottom of a grade 4 is the same as the bottom of a grade C. the bottom of a 7 is the same as the bottom of an A, and the bottom of a G is the same as the bottom of a 1.
 Equivalent outcomes – grade boundaries are being set so similar proportions of students achieve each grade as in previous years
 Our marking is internally consistent
 Our students are likely to achieve similar outcomes to previous years.
Methods
Ranking
Put all the grades in descending order
Find the attainment percentages for last year in your subject (https://schoolsweek.co.uk/gcseresults2017uksubjecttables/ for national data)
Use the previous attainment data to work out how many students are likely to achieve an A grade or better.
Count down from the top until you reach that number
The mark that student achieved is a good candidate mark for the Grade 7 boundary.
Repeat for the number of student achieving a C or better, to get the Grade 4 boundary.
Repeat again for the number of students achieving a grade, this would be the Grade 1 boundary.
You can then look at individual grade proportions to split out the 2,3,5,6 and 9 boundaries, or make a logical spread throughout the marks. e.g. if you find that the Grade 4 was at 30 marks and the Grade 7 was at 45 marks, you could put the 6 in at 40 and the Grade 5 at 35.
Some points to consider
This is not very far removed from how the exam boards will set their grades. They will just have a lot more numbers to play with.
There is no such thing as a completely accurate grade boundary. The best you can do is get a good idea of where a student is compared to the rest of the school and to the exact exam paper.
Grades generated this way would be projections, ie the grade that the test indicates they will achieve in the real exam. To convert to an actual Working At Grade you will need to raise the boundaries slightly.
Use old style exam boundaries
The old exam boundaries are likely to be significantly higher than the new ones due to the changes in challenge presented by the new courses.
For example, in Edexcel Mathematics
Grade 
2016 
2017 
Higher 
Foundation 
Higher 
Foundation 
A/7 
70% 

52% 

C/4 
35% 
71% 
17% 
51% 
G/1 

31% 

11% 
English is not as easy to compare as that has shifted from tiered to a single paper, but by looking at the single nontiered paper 3;
Grade 
2016 
2017 
A/7 
90% 
65% 
C/4 
59% 
45% 
G/1 
13% 
10% 
Typically, the percentage of marks needed to achieve a grade have shifted down by over 15 percentage point at most grades.
This is a huge difference compared to the normal drift of grade boundaries shifting each year. For maths, this has been in a range of about 6 percentage points (eg a C in maths has had a range from 29% to 35% from 2014 to 2016).
This being the case, shifting grade boundaries down by something in the region of 10 percentage points would not be unreasonable.
Make up grade boundaries based on gut feeling
Don’t
In some cases, this may seem like all you can do, but in reality you will effectively be using a ranking system without necessarily doing it formally. The advantage of doing it formally is that you can discuss how you got to the grade without having to say you made it up.
Hybrid
 Set grade boundaries using an adjustment (+/ fudge factor) on previous grade boundaries.
 Rank order the students and see if the grades look right; ie the students are getting the sort of grades you expect based on their work in lesson.
 Adjust the boundaries until the results look right. NB this is not adjust the boundaries until the results look good
 Review the boundaries and makes sure they are not too far away from what has happened previously
A hybrid method is probably the best approach as it makes use of all the sources of information and combines them to give you something consistent and explicable.
Worked Example for Physics
Old Style GCSE grade boundaries (converted to number grades)
Grade 
Old GCSE % 
8 
75 
7 
60 
6 
45 
5 
38 
4 
31 
3 
17 
We are pretty sure they are too high, but how much too high?
An adjustment of 10% seems reasonable based on the maths data
Grade 
Old GCSE % 
Adjusted 
8 
75 
65 
7 
60 
50 
6 
45 
35 
5 
38 
28 
4 
31 
21 
3 
17 
7 
So looking at historical grade distribution for our school (we want to do better than this, but it’s a good place to start)
We should have in the region of

Historical % of students achieving that grade 
Number of Students in current cohort 
8 
10 
4 
7 
22 
9 
6 
30 
13 
4 
28 
12 
3 
10 
4 
Running through the results in order, the top 4 students should get an 8. The lowest mark from the top 4 students was 60%. So that makes a good grade boundary for the grade 8. Then another 9 students takes us to 42%, so a good boundary for the 7. And so on to find suitable boundaries for the rest of the grades.
Grade 
Old GCSE 
based on similar proportions 
8 
75 
60 
7 
60 
40 
6 
45 
33 
5 
38 
30 
4 
31 
20 
3 
17 
17 
Taking all those together we can derive an reasonable ‘final’ grade boundary based on both approaches.
Grade 
Old GCSE 
based on similar proportions 
Adjusted Old GCSE 
Final 
8 
75 
60 
65 
65 
7 
60 
40 
50 
45 
6 
45 
33 
35 
37 
5 
38 
30 
28 
33 
4 
31 
20 
21 
25 
3 
17 
17 
7 
15 
Plugging these into the class results gives a set of data that looks about right, i.e. the distribution is slightly worse than students achieved last year, so we are probably not making it too easy, but they are not so much worse that we can’t work out who to target.
It is important to remember the point behind grading tests at this stage – we need to work out who we need to work with to get help them achieve their potential.