[ < ] | [ > ] | [ << ] | [ Up ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |

Acording to previous section we can consider an observation to be
controlled if its coefficient *f > 0.1*. Any controlled
observation can be eliminated from the network without corrupting the
network consistency—network reduced by one controlled observation can
be adjusted and all unknown parameters can be compute without the
eliminated observation.

Estimate of real error of *i*-th observation is defined as

(9) e_li = L^red_i - l_i, |

where
e_li
is value of *i*-th observation and
is value of *i*-th network element computed from adjusted
coordinates and/or orientations of the reduced network. Similarly is
defined the estimate of real error of a residual

(10) e_vi = L^red_i - L_l. |

Adjustment results are the best statistical estimate of unknown
parameters that we have. This holds true even for adjustment of
*reduced* network which is not influenced by real error of
*i*-th observation. On favourable occasions differences (9) and
(10) can help to detect blunders but to interpret these estimates as
*real errors* is possible only with substantial exaggeration.
These estimates fail when there are more than one significant
observational error. Generally holds tha the weaker the element is
controlled in netowrk the less reliable these estimates are.

Estimate of real error of an observation computes program `gama-local`

as

e_li = v_i/(p_i q_vi) |

and estimate of real error of a residual as

e_vi = e_li - v_i. |

[ < ] | [ > ] | [ << ] | [ Up ] | [ >> ] |

This document was generated by *Ales Cepek* on *June 26, 2017* using *texi2html 1.82*.