INTRODUCTION
In an earlier thread I was reading, there was a discussion of whether or not to use professional cavalry or stick with feudal cavalry during the Late period. Arguments against professional cavalry (gendarmes) were made on the grounds that gendarmes' stats are nearly identical to that of men-at-arms except for their higher upkeep cost. This is in fact the case; however, wondering if CA would really make a redundant unit that provides no incentive for use -- it takes a time and effort to make nice-looking meshes, which the graphics folks wouldn't want to see wasted -- I decided to run an experiment to see if gendarmes are actually inferior to feudal cavalry. Reasoning that any cavalry inferior to men-at-arms will be inferior to the Knights Templar, I set gendarmes and Templars against each other to see what happened.
THE EXPERIMENT
MAP: Grassy Plain
DIFFICULTY: Medium
FLORINS: 10,000
PERIOD: All
FRANCE: 1 unit of gendarmes
PAPAL STATES: 1 unit of Knights Templar
NUMBER OF TRIALS: 20
PURPOSE: To determine whether the gendarmes of professional armies are comparable in quality to the Knights Templar, representing the pinnacle of pre-professional heavy cavalry.
METHODOLOGY: Blocked experiment. I played trials 1-10 as France and trials 11-20 as the Papal States, to control for human advantage. In each battle, I single-clicked on the opposing unit and hit the fast forward button, not touching the keyboard again until the battle was over. After each battle I recorded how many soldiers each side lost, how many they killed, and whether or not the gendarmes won. Neither the gendarmes nor the Knights Templar were upgraded with weapons, armor, or experience. Statistics were computed by entering the results of the battles as arguments for MATLAB functions. The reason why I blocked the experiment comes from an earlier experiment where I did the same thing with both armies consisting of identical units of Gendarmes, and I beat the computer 16 of 20 trials despite the fact that all I did in each battle was click once and wait. The likelihood of that happening if my controlling the army made no difference is about 1 in 216.
~~~~~~~~~
The following are basic statistical assessments of the outcome of the experiment.
GENDARMES:
Victories: 60%
Casualties
Mean: 29.850
Standard Deviation: 9.0134
Minimum: 14
First Quartile: 23
Median: 29
Third Quartile: 39
Maximum: 41
Interquartile Range: 27
Kills
Mean: 35.80
Standard Deviation: 6.4531
Minimum: 18
First Quartile: 36
Median: 40
Third Quartile: 40
Maximum: 41
Interquartile Range: 23
KNIGHTS TEMPLAR:
Victories: 40%
Casualties
Mean: 33.950
Standard Deviation: 7.7083
Minimum: 15
First Quartile: 31
Median: 37
Third Quartile: 40
Maximum: 41
Interquartile Range: 26
Kills
Mean: 32.350
Standard Deviation: 7.78849
Minimum: 16
First Quartile: 27
Median: 34
Third Quartile: 39
Maximum: 41
Interquartile Range: 25
~~~~~~~~~~~
The following are data from our control group, where French gendarmes fight Spanish gendarmes with the same experimental procedure as when the gendarmes were fighting Knights Templar.
FRANCE
Victories: 60%
Casualties
Mean: 25.90
Standard Deviation: 10.305338
Minimum: 10
First Quartile: 17
Median: 29
Third Quartile: 33
Maximum: 41
Interquartile Range: 31
Kills
Mean: 36.850
Standard Deviation: 4.19618
Minimum: 28
First Quartile: 37
Median: 38
Third Quartile: 40
Maximum: 41
Interquartile Range: 13
SPAIN
Victories: 40%
Casualties
Mean: 36.05
Standard Deviation: 5.45291
Minimum: 24
First Quartile: 34
Median: 39
Third Quartile: 39
Maximum: 41
Interquartile Range: 17
Kills
Mean: 29.0
Standard Deviation: 10.09429
Minimum: 12
First Quartile: 20
Median: 34
Third Quartile: 39
Maximum: 41
Interquartile Range: 29
~~~~~~~~~~~
To test our hypothesis that the gendarmes do not perform differently than Knights Templar, let us average the casualty and kill rates of the gendarmes, assume that the performance against Templars was the same as against gendarmes, and find the z-scores of the observed casualty and kill rates.
(mean control kills) = 32.925
(std. dv. control kills) = 7.1452
(mean control casualties) = 30.975
(std. dv. control casualties) = 7.8791
(z score) = (mean - observed)/(std. dv.)
(z score observed kills) = -0.40237
(z score observed casualties) = .14278
Z scores of absolute value less than 0.5 are nowhere near strong enough for us to reject the null hypothesis. Therefore, to the best of my present ability to determine, there is no reason to claim that gendarmes are more or less effective in battle than Knights Templar. Given that they're cheaper to recruit, don't require Guild buildings, and the high upkeep cost is counterbalanced by high income from cities, gendarmes seem like a very good deal all around.
CONCLUSIONS
Why gendarmes come out about equal with Knights Templar is a matter best left to more experienced players than myself -- my knowledge of the Total War engine is not expansive. I most certainly believe they should not be dismissed as jumped up men-at-arms; similar experiments pitting gendarmes against men-at-arms saw gendarmes win 90% of the time with far fewer casualties and far more losses inflicted. For some reason or another, the unit card simply doesn't tell the whole story.
Bookmarks