The most typical concept of "going to scale" is expanding one program model to cover all the children in a target population, generally all children of a certain age or all caregivers of these children. An alternative vision of scale, which could be called scale through association, is achieved by piecing together several models, each of which might be small or medium-sized, each initiated or carried out with relative independence, and each approaching the problem in its own way but with a common goal or framework (Myers 1992b, chapter 14).

Scale by expansion
By focusing on one model, scale by expansion makes it easier to administer programs in a centralized system and to calculate coverage. But it does not make it easy to respond to variations in cultural, economic, and geographic contexts. Instead, imposing one model makes it difficult to adjust programs to local conditions and to achieve local participation. Variants on the model are needed. Moreover, although trying to provide care to all children may reflect idealism or a politically useful hope, not all children need the support of official programs, and some children need that support much more than others.

Expanding one approach or model or treatment may be effective in seeking scale when the object is to immunize children. Early childhood development does not lend itself easily to an immunization approach, however. But if the concept of scale for ECCD projects and programs is to expand one model, experience suggests that:

• Gradual extension is better than instantaneous extension.

• Establishing a core of trained people with experience working in the pilot or demonstration project helps by firmly grounding knowledge as the program expands. And projects that have worked well in particular areas can be training grounds for new people.

• Programs need to be adjusted as they expand, to reflect the lessons of experience. A large-scale version of a program usually will not work the same way as a smaller pilot or demonstration project based on the same model. To ease the adjustments, rigid requirements should be avoided.

There is a tendency to try to mount large-scale programs rapidly, using one model everywhere in a country and with little attention to the need for local adjustments. This tendency toward scale by explosion is linked sometimes to the pressure imposed by political cycles and desires, sometimes to a desire to move money made available through budget allocations or loan agreements. A strategy based on "bigger and faster is better" can be very positive in giving early childhood development broad exposure and getting activities under way in many places. And there is a natural desire to take advantage of political will and budgets while they last. But such a strategy is likely to produce programs of low quality, unadjusted to local conditions and with little local participation. In short, important tradeoffs must be taken into account when seeking scale with one model. Program designers and managers must seek ways of moderating the possible negative effects of pressures to go to scale in a hurry.

Scale by association
Scale by association can be visualized as the process of putting together the pieces of a puzzle. The puzzle has a framework. Each piece may be a different shape, and the colors will differ too, depending on the place the piece occupies in the puzzle. When all the pieces are in place, a complete picture will appear and scale will have been achieved. The more pieces in place at one time, the greater the scale.

Putting the pieces together must be a collaborative process, with some general rules to guide it and some incentives to help the picture emerge more rapidly. If there is no clear framework to provide general guidelines and delimitations—an ECCD strategy—efforts may overlap at the same time that some groups are left aside. But if there is such a strategy, scale by association allows room for real local participation in the choice and operation of ECCD models appropriate to local conditions, something much harder to achieve in extending one predetermined model.