The data, thirty years on

A correction the spreadsheet makes in the chapter's favour. The thesis reports the plot ranking exactly right — site four highest, then two, three, six, five, one — but then concludes the Deep River result "may be skewed," because "the two sites with the next biggest trees were in Monkey River." They were not. M2 and M3 are Deep River 5 and Deep River 4. (The M codes look like they run in order of establishment rather than by river, which would explain the slip — but that is an inference; nothing in the files states what the prefix means.) All three Deep River plots outrank all three Monkey River plots on girth, diameter, mass and height. Reading the M prefix as "Monkey" cost the chapter its clearest result.

What "no overlap" means here — and what it does not. The separation is between the six plot means, on all four size metrics. Individual stems overlap freely: Monkey River 1 holds a stem of 113.0 cm girth, larger than every Deep River plot mean, and the Deep River plots run down to 8.5 cm. With three plots per watershed and seventeen stems at Deep River, a clean separation of six means is suggestive, not conclusive; the chapter's own nested ANOVA is the right strength of claim.

Two labels that need care. The chapter calls this the "Suva Barrier Reef." In the modern literature that name belongs to a different structure, on the harbour side of the peninsula; the reef surveyed here is part of the Laucala Barrier Reef system — Sosoikula, Nukubuco, Nukulau and Makaluva. That identification is itself an inference from the chapter's description, not a mapped fact, and the site map has not yet been recovered. The naming is worked through on the Fiji return page. The fieldwork year is not recorded either: Methods give only "mid August to mid October on Thursday mornings," which puts the survey in the austral spring of 1993, 1994 or 1995, with 1994 the most likely.

The English wording is not quite the instrument people answered. The item labels charted and tabulated here are the English ones carried in the SPSS file, but about 96% of the 260 respondents filled in a Latvian questionnaire that is not a clean translation of it. The Latvian anchors read ļoti patiess un pareizs / pareizs / nepareizs — "very true and correct / correct / incorrect" — a correctness frame rather than a gradation of truth, with the "somewhat" steps gone. Several items diverge outright. q24, printed above as "There are actions that I could take which would be effective in helping the environment" (mean 2.31), asks in Latvian Man ir tieši ieteikumi kas varētu palīdzēt vides aizsardzības jomā — closer to "I have specific suggestions that could help." A low score there may record modesty about having proposals rather than a belief that acting is futile. The Latvian text is the operative instrument; read the English labels as approximations. The item-by-item comparison is on the Latvia return page.