Manifesto Coding
- Introduced by
- Manifesto Research Group / Comparative Manifesto Project (CMP/MARPOR)
- Year
- 2001
- Type
- Quantitative content analysis of party manifestos
- Family
- Content analysis / text coding
What does it do?
Manifesto coding is the quantitative content-analysis methodology of the Comparative Manifesto Project (CMP/MARPOR) for measuring parties' policy preferences from their election manifestos.
Trained coders break each manifesto into quasi-sentences and assign every unit to one of a fixed set of policy categories. Counting how often each category appears yields salience measures, and combining pro- and anti- categories produces position scores such as the left–right RILE index, giving comparable estimates of party positions across more than fifty democracies since 1945.
The intuition behind the method
A party manifesto is a sales pitch in which a party chooses how much to emphasize each issue. Manifesto coding rests on the saliency theory of party competition: parties compete less by taking opposite stands on the same issue than by stressing the issues they own. So instead of asking 'does the party favor or oppose X,' coders count how much attention the manifesto devotes to each of many predefined topics. The mix of emphases — lots of free-market talk versus lots of welfare talk — locates the party on policy dimensions, and the resulting numbers can be compared across countries and decades.
History
The Manifesto Research Group began coding party manifestos in the late 1970s, and the project matured into the Comparative Manifesto Project. Budge, Klingemann, and colleagues' 2001 volume Mapping Policy Preferences published the methodology and the first comprehensive estimates for parties, electors, and governments from 1945 to 1998. The 2013 follow-up, Mapping Policy Preferences from Texts, addressed statistical and validity critiques, and Lowe and colleagues' 2011 paper proposed improved scaling of the coded categories. Now run as MARPOR, the project remains a central data infrastructure for comparative party research.
When to use it
Use manifesto coding when you need comparable, long-run measures of party policy positions and issue emphases grounded in parties' own authoritative statements, especially for cross-national and over-time analysis of party competition, government policy, and representation. The CMP/MARPOR dataset is the standard source. It is less appropriate when you need positions on a specific contemporary issue not captured by the fixed scheme, when manifestos are unavailable or unrepresentative of a party's true stance, or when the salience-based logic does not match a confrontational issue dimension; expert surveys or text-scaling may then fit better.
Limitations and considerations
- Human coding is labor-intensive and subject to inter-coder unreliability, particularly across languages and over time.
- The saliency-based RILE index conflates emphasis with direction and has contested measurement properties.
- The fixed category scheme can miss newly salient issues and forces diverse content into predefined boxes.
- Manifestos may not reflect a party's revealed behavior in government, limiting validity as a preference measure.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Provides the longest-running, most comprehensive comparable dataset of party positions across democracies since 1945.
- Grounded in parties' own authoritative documents rather than perceptions, giving a direct behavioral measure of emphasis.
- A fixed, theory-driven category scheme enables consistent cross-national and over-time comparison.
- Transparent and reproducible: raw category counts are published, so alternative scales and reanalyses are possible.
Applications
- Estimating party left–right positions across countries and elections for studies of party competition.
- Measuring issue salience and agenda emphasis in party platforms over time.
- Analyzing congruence between party, voter, and government policy positions in representation research.
- Validating and benchmarking automated text-scaling methods such as Wordscores and Wordfish.
- Tracking the emergence and decline of policy dimensions, such as environmental or European-integration issues.
How it works
How it works
Each manifesto is divided into quasi-sentences, the unit of analysis: a quasi-sentence is an argument or statement expressing a single political idea, which may be a full sentence or part of one. Consistent unitization is the foundation of reliability, since every later count depends on how the text was segmented.
Trained coders assign each quasi-sentence to exactly one of the project's standard policy categories (the scheme has around fifty-six, grouped into domains such as economy, welfare, and external relations). Coding follows a detailed handbook to maximize cross-coder and cross-country consistency, and coder training and reliability checks are integral.
For each category, the count of quasi-sentences is divided by the total number of coded quasi-sentences in the manifesto and multiplied by one hundred, giving the percentage of the document devoted to that category. These salience shares are the raw, comparable measures the project distributes.
Formula
s_{ck} = \frac{n_{ck}}{N_k} \times 100Position indices combine category saliences. The canonical left–right RILE index sums a fixed set of right-emphasis categories and subtracts a fixed set of left-emphasis categories, producing a single left–right score per manifesto. Domain-specific scales are built analogously from relevant categories.
Formula
\text{RILE}_k = \sum_{c \in R} s_{ck} - \sum_{c \in L} s_{ck}Because RILE's additive subtraction has known weaknesses, alternative scalings exist. Lowe, Benoit, Mikhaylov, and Laver propose a logit-style ratio scale that takes the log of the ratio of right to left emphasis (with a smoothing constant), which has better measurement properties and a clearer interpretation as a position rather than a salience difference.
Formula
\theta_k = \log\!\frac{R_k + 0.5}{L_k + 0.5}Common pitfalls
- Treating RILE scores as interval-level positions when their construction mixes salience and direction, distorting comparisons.
- Ignoring coder reliability and the resulting measurement error, which can be substantial for individual manifestos.
- Comparing raw category counts across manifestos of very different lengths without using the salience percentages.
- Assuming the standard category scheme captures an issue (e.g., a new cultural cleavage) that it was not designed to measure.
Frequently asked questions
What is a quasi-sentence and why does it matter?
A quasi-sentence is the project's unit of analysis: a statement expressing a single political idea, which may coincide with a grammatical sentence or be a clause within one when a sentence makes several distinct points. It matters because every salience and position measure is a count of quasi-sentences, so inconsistent segmentation across coders or languages introduces measurement error before any category is assigned. Reliable unitization is therefore a prerequisite for valid manifesto data.
What are the main criticisms of the RILE left–right index?
RILE sums right-emphasis categories and subtracts left-emphasis ones, which mixes how much a party talks about an issue (salience) with which side it takes (position), and treats the two directions symmetrically. Critics show this can make scores unstable and sensitive to the fixed category list, and that it behaves more like a salience difference than a true spatial position. Alternatives such as the logit ratio scale of Lowe and colleagues address some of these measurement concerns.
How does manifesto coding compare to automated text scaling?
Manifesto coding is human content analysis using a fixed, theory-driven category scheme, yielding interpretable, comparable category counts but at high cost and with inter-coder reliability concerns. Automated scaling methods like Wordscores and Wordfish estimate positions from word frequencies far more cheaply and are often validated against manifesto data. The approaches are complementary: manifesto codings provide a rich, validated benchmark, while automated methods scale to large corpora and new texts.
Related methods
Sources
Budge, I., Klingemann, H.-D., Volkens, A., Bara, J., & Tanenbaum, E. (2001). Mapping Policy Preferences: Estimates for Parties, Electors, and Governments 1945–1998. Oxford: Oxford University Press.
ISBN: 9780199244003Copied.Volkens, A., Bara, J., Budge, I., McDonald, M. D., & Klingemann, H.-D. (Eds.) (2013). Mapping Policy Preferences from Texts: Statistical Solutions for Manifesto Analysts. Oxford: Oxford University Press.
ISBN: 9780199640041Copied.Lowe, W., Benoit, K., Mikhaylov, S., & Laver, M. (2011). Scaling Policy Preferences from Coded Political Texts. Legislative Studies Quarterly, 36(1), 123–155.
Open DOI referenceCopied.