Cumulative search error

The question

Next stop on the tour, and it's a busy one. How do we score a search fairly when every swim starts from a different place? You're watching Morris Water Maze Software Guide, over at morriswatermazesoftware.com.

Meet the cumulative search error

Cumulative search error, from the same 1993 work as the proximity measure, adds up the animal's distance from the platform over the whole trial. To be fair to far start points, it subtracts the distance an ideal swimmer would have built up on a direct path at the same speed. What's left is the error that comes from searching in the wrong place.

How it runs

Here's how it plays out. First, sample distance to the platform at fixed intervals through the trial. Next, sum those distances to get the total. Then, work out the sum an ideal direct swim from the same start would produce. Finally, subtract the ideal sum from the actual sum to get the cumulative search error.

Strengths and limits

So what's in it for you? It corrects for start position. It combines speed, direction and distance into one score. It's sensitive to small but real differences between animals. Where does it bite? It needs a steady sampling rate and good tracking. It's harder to explain to readers than latency.

What to measure

What should you actually measure? Cumulative search error per trial: the main score; lower is better. Search error across days: a learning curve that isn't fooled by speed. Proximity: the averaged cousin, often reported alongside. Path efficiency: a quick check on how direct the path was. A fast rat from a far start and a slow rat from a near start can share the same latency. Cumulative search error removes the start advantage and shows which one actually headed for the goal.

The takeaway

Short version: search error across days and its neighbours tell you how the task was done, not just how fast. The full transcript is on this page, with sources.