Presentation is loading. Please wait.

Presentation is loading. Please wait.

Quasi-experimental Evaluation Design for the NECT Learning Programme

Similar presentations


Presentation on theme: "Quasi-experimental Evaluation Design for the NECT Learning Programme"— Presentation transcript:

1 Quasi-experimental Evaluation Design for the NECT Learning Programme
SAMEA Conference 27 October 2017

2 Background to the NECT Learning Programme

3 Background to the NECT In line with the NDP, the Education Collaboration Framework is a partnership between government and key stakeholders (business, labour and civil society) to increase collaboration and improve outcomes The National Education Collaboration Trust (NECT) was established to operationalise the framework and oversee the implementation of programmes The District Improvement Programme (DIP) is the NECT’s flagship programme and accounts for the vast majority of its expenditure The DIP leverages more than R160m per annum and is implemented across all schools in 8 districts in the country (and growing)

4 What is the District Improvement Programme?
The Learning Programme is the “common core” of the DIP; focussed within these components Teacher Professionalisation Standardised programmes, content knowledge and fundamentals of performance. Courageous and Effective Leadership Resources management, curriculum tracking and support. Courageous leadership module as running thread Capacity of the State to Deliver Training and coaching to drive and maintain education change. System level innovations Learner Welfare and Support Psycho-social health assessment test. Reigniting the role of Council of Learners in education and development. Parent and Community Participation Community members , unions, business member District Steering Committee overseeing education improvement in 8 districts. Volunteerism in schools. Source: NECT (2016)

5 What is the Learning Programme?
The Learning Programme has three key levers: Detailed Lesson Plans and Curriculum Trackers High Quality Materials Instructional Training and Coaching The key focus of the Learning Programme is to improve: Curriculum management and oversight Curriculum coverage Assessment practices

6 Evaluation methodology

7 Evaluation Methodology: Overview
Impact evaluation Focussed on Foundation Phase teaching and learning EGRA and EGMA type learner assessment tools in 150 schools Grade 1 assessment in early 2017 and Grade 2 end of 2018 Supplement by use of learner workbook data Supported by a process evaluation Semi-structured interviews at national district and school level In-school data collected by evaluation team NECT monitoring data Case studies in a sub-set of treatment schools

8 Choosing a quasi-experimental approach
Key to impact evaluation is establishing a credible counterfactual The big challenge is to remove (or at least minimize) unobserved biases An RCT attempts to remove bias by randomising treatment in advance However, the Learning Programme started in 2015, while the evaluation only started in 2017 Randomisation therefore not possible A quasi-experimental approach was therefore selected: matching

9 What is matching? Matching involves constructing a comparison group that is sufficiently similar to the treatment group at the outset of the programme based on observable characteristics In contrast to an RCT, this comparison group is typically selected ex-post Since the programme is mainly administered at school level, we select comparison schools that are sufficiently similar to a set of sampled treatment schools that received the programme; using DBE data available on these schools School level because the programme is implemented at this level

10 Approach to matching Initially considered choosing comparison schools from within treatment districts: However programme was expanded across all schools in treatment districts (universalisation) As it is a district level programme, there would also have been within district spill-overs through e.g. district office intervention This meant that schools had to be selected from other districts Matching would thus involve: First selecting comparison districts Then selecting comparison schools from within comparison district School level because the programme is implemented at this level

11 Selecting comparison districts

12 Focussed District Model
District selection Focussed District Model Whole District Model Vhembe The Learning Programme comprises two different implementation models: The Whole District Model (WDM) The Focussed District Model (FDM) Programme is implemented by 3 different lead agents One WDM district and two FDM districts used as treatment districts based on a number of criteria (next slide) First step is looking at potential comparison district for each treatment district Waterberg Bohlabela Bojanala B uThungulu U P Pinetown M Mt Frere L Libode The Learning Programme c Source: DBE (2013) Education district in South Africa: A review

13 District selection - factors
Tshipisi Sagole Mogala-kwena Mopani Vhembe Waterberg Availability of appropriate comparison districts: School profile and performance Language profile of district Socio-economic similarities Limited spill-overs and other programmes One district from each lead agent Composition of treatment district: Relatively homogenous language profile Quintile distribution of schools in the district Lebowa-kgomo Bohlabela Sekhukhune Ngaka Modiri Molema Bojanala Ehlanzeni Dr KK Umkhanyakude U Zululand Uthungulu Umgungundlovu Ilembe P Pinetown U Umlazi Mt Frere M L Mbizana Libode Lusikisiki Source: DBE (2013) Education district in South Africa: A review

14 Example 1: Bohlabela (language)
Mopani No good language matches in Mpumalanga Mopani is in Limpopo – cross province comparison not ideal Hence no good matches Bohlabela Ehlanzeni Umkhanyakude Source: DBE (2013) Education district in South Africa: A review

15 Example 2:Vhembe (spill-overs)
Tshipisi Sagole Mogala-kwena Vhembe Waterberg Tshipise Sagole (TS) relatively similar to Vhembe across language and quintiles However project managers reported significant spill-overs between districts Used to be a single district District management still integrated Project materials have been used in TS No other district has sufficient Tshivenda schools to act as counterfactual for Vhembe Lebowa-kgomo Sekhukhune Source: DBE (2013) Education district in South Africa: A review

16 Example 3: Eastern Cape Districts broadly similar across key characteristics Mt Frere seen as having stronger implementation than Libode Mbizana selected as comparison district for Mt Frere Mt Frere M L Mbizana Libode Lusikisiki Source: DBE (2013) Education district in South Africa: A review

17 Districts selected Treatment district Comparison district Language
Bojanala Dr Kenneth Kaunda Setswana Mount Frere Mbizana isiXhosa Uthungulu Ilembe isiZulu Bojanala Dr KK P Pinetown U Umlazi Mt Frere L Lusikisiki Source: DBE (2013) Education district in South Africa: A review

18 Selecting comparison schools

19 Matching: School level (1)
The matching process was to: Select a random sample of treatment schools Select matching comparison schools based on this sample Ideally, comparison schools should be selected using the criteria used to select schools into the programme. However programme selected was not based on objective criteria that could be replicated using available data

20 Matching: School level (2)
The data fields used were selected based on three main factors: Data must be available across all schools Fields (such as quintile) that were seen as good proxies for the original programme selection criteria Fields that are indicative of school performance at the baseline of the programme (2014)

21 Matching methodology (1)
A variety of different matching approach exist, e.g.: Propensity Score Matching (PSM) Exact Matching Matching based on distance metrics PSM regressions yielded few statistically significant parameter estimates and a very narrow band of propensity scores Hence a hybrid approach between exact matching and Euclidean distance matching was used

22 Matching methodology (2)
In particular, exact matching was performed to stratify schools into groups across the following fields: School quintile (1, 2 or 3) School type (Primary or Combined) Grade 1 enrolments, grouped into 4 categories Within the above categories, nearest neighbour matching was then performed by minimising the Euclidean distance measure of each school’s average 2014 Annual National Assessments marks across Grade 1 and 3 mathematics and language

23 Thank you!

24


Download ppt "Quasi-experimental Evaluation Design for the NECT Learning Programme"

Similar presentations


Ads by Google