htna-named wrapper for Nestimate::markov_order_test(). Tests
whether a first-order Markov model is adequate for the observed
sequences, or whether a higher order is required, by comparing the
empirical transition structure against orders 1..max_order via
permutation.
Arguments
- data
An htna network from
build_htna(), or raw sequence data: a list of character vectors or a wide-format data frame. SeeNestimate::markov_order_test().- ...
Additional arguments passed to
Nestimate::markov_order_test(), such asmax_order,n_perm,alpha,parallel,n_cores, andseed. See that function for the full, current argument list.
Value
An object of class net_markov_order with components
optimal_order (sequential permutation-test order), bic_order,
aic_order, test_table (a data frame with one row per tested order
and columns order, loglik, AIC, BIC, df, g2,
p_permutation, p_asymptotic, significant), the permutation null
distributions, fitted transition matrices, and the inputs. print(),
summary() and plot() methods are supplied by Nestimate; see
Nestimate::markov_order_test().
Details
Accepts an htna network directly (its session sequences are tested),
or raw sequence data (a list of character vectors or a wide-format data
frame). Use alongside reliability_htna() and
casedrop_reliability_htna() when assessing whether a Markov-1
transition network is appropriate before trusting downstream htna
analyses.
The orders are tested sequentially: optimal_order is the highest order
k for which every step from order j - 1 to j, j = 1, ..., k, is
significant at alpha. It is therefore one less than the first order
whose step is not significant, and equals max_order when every step is
significant (in which case the search should be widened). The BIC- and
AIC-preferred orders are reported alongside.
Suffixed _htna to avoid clashing with
Nestimate::markov_order_test() when both packages are loaded.
Examples
data(human_ai)
net <- build_htna(human_ai, actor_type = "actor_type")
#> Metadata aggregated per session: ties resolved by first occurrence in 'session_date' (1 sessions), 'cluster' (42 sessions), 'actor_type' (24 sessions)
# Few permutations to keep the example fast; use the default 500 in practice.
markov_order_test_htna(net, max_order = 2, n_perm = 20, seed = 1)
#> Markov Order Test [within-w permutation, n_perm = 20, alpha = 0.050]
#> 428 sequences / 19346 observations / 12 states
#>
#> Selected order BIC: 1 AIC: 2 permutation-LRT: 2
#>
#> order loglik AIC BIC df g2 p_permutation p_asymptotic
#> 0 -43741.91 87505.83 87592.40 NA NA NA NA
#> 1 -37315.96 74899.92 75954.53 121 12815.99 0.04761905 0
#> 2 -33812.56 70683.12 82716.72 1272 6481.76 0.04761905 0
#> significant
#> NA
#> TRUE
#> TRUE
