Epi in the News

Thanks Sujit.

It is very concerning to see such high rates of school exclusion, in particular in the groups mentioned. I hope the findings of the study can be put to good use in helping more children stay in school and in fostering better understanding and trust between schools and children from all backgrounds.

1. The exclusion rate appears to be an incidence risk.
No. of children excluded in a school year/No. of children who start out in that school year.

2. The comparison groups are Black Carribean, Gypsy, Irish Traveller, Mixed race White, Roma and White children. The baseline group is White children.

3. (No. of Black Carribean children excluded in a school year/No. of Black Carribean children who start out in that school year)/(No. of White children excluded in a school year/No. of White children who start out in that school year).

4. Variance.

5. I think you're right Sujit - it doesn't make sense. I think they are trying to say that the Cambridge result may not be statistically significant because the sample size is low.

6. It would be difficult to design a study to determine whether exclusion had a long term impact on students' social and economic well being. The only method that would really deal with the significant problems of bias and confounding would be a randomised controlled trial where some students who were about to be excluded were instead randomised to stay at school. This might well not be feasible as my understanding is that children should only be excluded from school if the school considers it impossible to continue with them remaining on the school roll. Any other study would be beset by confounding.

Judith

I agree Judith.. the rates are concerning... this article represents the unpleasant reality of the world actually...

Coming to the epidemiology...

1) Here the exclusion rate is the number of exclusions divided by the total headcount in each group of students. Now, I was thinking the term used here is rate- might not be very accurate as there is no time in the denominator. Probably a better term would be risk. But then I realized that there is a time factor here- the numbers are for 1 academic year... so I guess its not so wrong to use the term rate here... The time factor is just 1 year.

2) The baseline group is the white children. The comparison groups include Carribean, Gypsy, Traveller, and Roma.

3) When they say "6 times higher", I am thinking this is the value of the rate ratio..
So the equation would be exclusion rate of Carribean children / exclusion rate in white children = 6

4) Hmm.. I dont know actually...

5) Agree... It doesn't make sense. Clearly, somebody who doesn't understand how the numbers or ratios should be interpreted or explained accurately has written the article... :-)))

6) Well... that would be a cohort study... follow the children who were excluded and observe the outcomes... I would only consider a retrospective cohort study (Identify children who were excluded from schools maybe 20-30 years ago.. and see what is their socioeconomic status today). The reason being anything prospective here would be unethical - be a randomized intervention trial or an observational cohort study. It would be unethical of course to design an RCT where the intervention would be school exclusion.. Even a prospective cohort would be questioned- how do you identify a child that has been excluded from a school, and not help improve the situation, especially if resources are available to do so.
And like Judith says there would be confounders to deal with.. but the results would be real life and valid.

Fathima

1. The numerator is the number of students excluded in an ethnic group in a particular school, in the 2018-2019 academic year. The denominator is the total number of students of the same ethnicity in the same school, during 2018-2019 academic year.

Based on the information available, I think this is an incidence risk because we are not given any detail on whether they accounted for students leaving the school during that time period in their calculations. It seems to me that the denominator is not the total person-time at risk during the time period but rather the number of people at risk at the start of the time period even though they have called it an exclusion "rate".

2. The main groups being compared in the Guardian analysis are white British and black Caribbean. Mixed-race white, Gypsy, Roma and Traveller children are also mentioned.
White British is the baseline group in this article.

3. It appears that they divided 27.1% (exclusion rate in black Caribbean students) by 4.1% (exclusion rate in white British students) which is a risk ratio

4. Unsure

5. The first sentence makes sense to me- seems to be just describing their results, although the wording could be improved.
The second sentence doesn’t really make sense; perhaps they wanted to say the sample size was small in Cambridgeshire and thus we should exercise caution when interpreting these results.

6. Ideally, I would like to do a cohort study. You could utilise local authority/school data to locate students who had been excluded (exposed group) and then your unexposed group could be students from the same schools who had not been excluded (unexposed group).
“Social and economic well-being” is quite a broad outcome so you would need to consider what your indicators were carefully. On a first pass, some possible indicators could be lack of criminal activity, university education, employment status, socioeconomic status, relationship/marital status.

 
Possible confounders would include ethnicity (as per this article it's associated with exclusion rates and unfortunately due to ongoing structural racism/discrimination it may also be associated with long-term social and economic well-being) and family socioeconomic status.

Depending on the time frame and available data, you could consider either a prospective cohort study or a historic cohort study. Naturally, a prospective cohort study will be more expensive and take longer and I agree with Fathima’s points that you would need to consider the ethics carefully, but if historic records are not complete with needed data (including possible confounders) it will not be possible to do a historic cohort.

Some limitations of this study design: loss to follow up may introduce bias to the results and, as Judith noted, there is always a possibility of residual confounding so it would be important to see if there was consistency of findings from other studies.