To anonymize learner data in Excel, work on a copy of the class list and delete every column the study does not use. Shuffle the rows, give each learner a code, and move the names and Learner Reference Numbers (LRNs) with their codes into a separate key file that stays in the school. Then turn birth dates into ages or age bands, and count how many learners share each remaining row until no row describes only one learner. DepEd Order No. 16, s. 2017, the Research Management Guidelines, asks researchers to hand in “an executive data set in an accessible file format (i.e. Microsoft Excel file)” with the completed study, but it never says to take the names out first. That duty follows from the Data Privacy Act’s rule against excessive data and DepEd’s rule that a learner’s identity “shall be kept confidential.”

Deleting names and LRNs is only the start. In a class of 18, a row that reads “Male, 12, Grade 5” may fit one learner only, and then anyone who knows the class knows whose scores it holds. The National Privacy Commission (NPC) treats coded data as personal data for as long as a key exists; only data that no one can trace back to a learner falls outside the law.

Key findings

  • DepEd asks for the data set but not for its anonymization. Researchers “will submit an executive data set” in Excel “for accountability and reproducibility of research”, and completed studies are archived “along with … the data sets used during the study” (DO 16, s. 2017, Secs. V.B.vii and V.B.ix). The order does not define the term, and its one amendment, DO 026, s. 2021, leaves both sentences as they were.
  • A class list’s scores are sensitive personal information. The Data Privacy Act lists age and education among sensitive personal information (RA 10173, Sec. 3(l)), and the NPC holds that a class roster’s grade level, section and test scores are sensitive because they relate to a learner’s education (Advisory Opinion No. 2020-046).
  • Coded is not anonymous. “Pseudonymized or coded information remains personal data and continues to be subject to the DPA,” and “the reidentification key should be separately secured” (NPC Advisory Opinion No. 2026-007).
  • Removing names and LRNs is not enough. Birth date, sex and a five-digit ZIP code “likely made” 87% of the 1990 United States population unique (Sweeney, 2000). In the invented class of 18 below, sex, exact age, grade and sitio leave 14 rows that fit one learner each; sex and grade alone leave none.
  • Hiding is not deleting. A hidden column, a hidden sheet, a PivotTable’s saved source data, a comment and the file’s author field all travel with the workbook, and the recipient can bring them back (Microsoft Support, Remove hidden data and personal information by inspecting documents).
  • The research exemption is narrow. The Act’s exemption for research applies “only to the minimum extent” necessary and “subject to the requirements of applicable laws, regulations, or ethical standards” (IRR of RA 10173, Sec. 5 and 5(c)). Consent is not always required, for example for research “where the end results will be anonymized and will only disclose the general demographic of the research subjects” (NPC Circular No. 2023-04, Sec. 16(C)(2)).
  • No learner’s personal data goes into a public AI tool. DepEd prohibits “the input, processing, or storage of personal and sensitive information within AI systems” in research (DO 003, s. 2026, Sec. VI.1.3).

What DepEd asks for: the executive data set

DepEd Order No. 16, s. 2017 sets the steps a study goes through, from the call for proposals to archiving. Under “Submission and Acceptance” it says: “In addition, for accountability and reproducibility of research, researchers will submit an executive data set in an accessible file format (i.e. Microsoft Excel file)” (Sec. V.B.vii). Under “Archival”, the Planning Service’s Policy Research and Development Division, “in close collaboration with” the regional Policy, Planning and Research Division and the division’s School Governance and Operations Division, “will set up mechanisms to archive all completed education researches, along with other relevant documents, such as the data sets used during the study” (Sec. V.B.ix).

That is all the order says about the data set. It does not define “executive data set” among its terms (Sec. III), its outline of a completed report has no part for data (Annex 6), and its tables of deliverables for each tranche of a Basic Education Research Fund (BERF) grant list a “Final Report” without a data file. Its ethics section asks for “ensuring confidentiality in handling data of respondents and providing consent forms where necessary” (Sec. VI.C), and points researchers to the Child Protection Policy until DepEd sets up a research ethics committee. DepEd Order No. 026, s. 2021, the order’s only amendment, adds online data gathering and some costs and leaves the data set and archiving sentences unchanged; its item 5 keeps every other provision “enforced unless repealed and amended by other DepEd Issuances.”

Two later issuances come closest to a rule on the data set:

  • DepEd Memorandum No. 028, s. 2022 gives the research committees a Quality Control Checklist for accepting BERF-funded research and archiving studies in E-Saliksik, used “Consistent with Section V-B-vii of the RMG.” For an action research report, items 10 and 11 say: “At the minimum, the completed research must comply with the provisions of the Data Privacy Act of 2012 (RA 10173), the DepEd Child Protection Policy (DO 40 s. 2012), and the National Indigenous Peoples’ Education Policy Framework (DO 62, s. 2011).”
  • DepEd Order No. 014, s. 2022, which set up E-Saliksik, DepEd’s research portal, keeps “consent forms, Memoranda of Agreement, pictures submitted as means of verification” out of the reports it uploads (Sec. VI.b). The portal archives research reports and the Planning Service’s public statistics; the order says nothing of a study’s data set.

The Action Research Toolkit: Teacher’s Guide (2019), made with DepEd’s research and teaching offices, says more about raw data than the orders do: “Raw data must be kept under lock and key and names of students must be replaced by code names or code numbers” (Sec. 4.1). It also warns that “some demographic data points might easily give away” the identities of respondents (Sec. 4.2).

Regional and division offices that collect the data set read it the same way and seldom add more. DepEd CALABARZON’s Regional Memorandum No. 158, s. 2023 tells BERF grantees, “with the purpose of better utilization of research findings,” that “the executive data set/gathered data must be in Excel format,” in a file named with the lead author’s surname, the division, the theme and the BERF cycle (items 1 and 2); it says nothing about names or LRNs. In the City Schools Division of Tayabas, the research committee’s secretariat checks “the executive data set if it reflects the results and findings in the study” (Division Memorandum No. 137, s. 2021, Sec. IV.F). One division writes the rule out: Bayugan City asks for an “Executive dataset (machine-readable, anonymized) and data dictionary,” and adds that “Identifiable data must be anonymized or pseudonymized prior to submission” (Division Memorandum No. 348, s. 2025, items 4 and 6).

DepEd’s national issuances do not define the data set, so what goes into it is a reading of the purpose DO 16 gives it, accountability and reproducibility. A committee checks a study’s numbers by rerunning its tables, and for that it needs the coded rows of the variables the analysis used and a codebook that says what each column and code means. It needs no name, LRN, birth date, address or phone number. That reading satisfies DO 16 and the Data Privacy Act at once, and it is what Bayugan City requires.

What the Data Privacy Act and DepEd’s rules protect

The Data Privacy Act of 2012 defines personal information as information “from which the identity of an individual is apparent or can be reasonably and directly ascertained by the entity holding the information, or when put together with other information would directly and certainly identify an individual” (Republic Act No. 10173, Sec. 3(g)). The last clause is the one that matters for a data set without names: a row that identifies a learner when put together with what classmates, parents and teachers already know is still personal information.

Much of a class list is sensitive personal information, which the Act guards more closely. Its list includes an individual’s “race, ethnic origin, marital status, age, color, and religious, philosophical or political affiliations” and their “health, education, genetic or sexual life” (Sec. 3(l)(1) and (2)). The NPC applies this to schools: “Apart from the student’s name, the data set enumerated included in the class roster, i.e. student’s school name, grade level, section and test scores, are considered sensitive personal information as these are related to the student’s education” (Advisory Opinion No. 2020-046).

DepEd’s own rule on the LRN goes further: “The identity or other information that may reasonably identify the pupil, student or learner shall be kept confidential” (DepEd Order No. 22, s. 2012, item 9). The enrollment policy adds that learners’ personally identifiable information “shall be used solely for the specific purpose for which it was obtained, and shall not be disclosed or disseminated to third parties … without the explicit consent of the learner and their parents or legal guardian, unless required by law or necessary for the legitimate purposes of the data controller” (DepEd Order No. 017, s. 2025, Sec. X.C). Access within DepEd is “limited to authorized personnel … on a need-to-know basis” (Sec. X.B).

What the research exemption covers

The Act does not apply to “Personal information processed for journalistic, artistic, literary or research purposes” (Sec. 4(d)), and teachers sometimes read this as leaving research data outside the law. Its rules narrow it three ways:

  • Only as far as the research needs. The exemption applies “only to the minimum extent of collection, access, use, disclosure or other processing necessary to the purpose,” and only to research “intended for a public benefit, subject to the requirements of applicable laws, regulations, or ethical standards” (IRR of RA 10173, Sec. 5 and 5(c)).
  • Security duties stay. The same section says the exemption does not reach controllers “who remain subject to the requirements of implementing security measures for personal data protection,” and Sec. 6 puts the burden of proving an exemption on whoever claims it.
  • Ethics first. The NPC has held that research which skipped the ethical standards that apply “cannot be considered as a special case under the DPA” (Advisory Opinion No. 2021-023).

The NPC’s consent guidelines name two cases where research does not always need consent; the one that fits a class study reads: “The conduct of research where the end results will be anonymized and will only disclose the general demographic of the research subjects does not require the consent of the data subject” (NPC Circular No. 2023-04, Sec. 16(C)(2)). That speaks of the end results, not the working file. A teacher still holds names until the key is cut, and parents’ consent (DO 16, s. 2017, Sec. VI.C) and the learner’s assent that the national ethics guidelines require (NEGRIHP 2022, social research item 16) still apply (DepEd Action Research Ethics: Consent, Assent and Privacy).

Research ethics asks the same. The National Ethical Guidelines for Research Involving Human Participants 2022 (NEGRIHP), issued by the Philippine Health Research Ethics Board for all research with human participants, say: “Records that link individuals to specific personal information shall not be released” (item 29). Its chapter on social research asks researchers to “maintain the full confidentiality of all information and the anonymity of participants by instituting the necessary protection procedures in all research materials” (item 24), and against the risk of stigma its table of protections names “pseudonymization or de-identification of materials and data at the soonest possible time” (Table 2).

Once data no longer identify anyone, the time limits ease: “Personal data which is aggregated or kept in a form which does not permit identification of data subjects may be kept longer than necessary for the declared, specified, and legitimate purpose” (IRR, Sec. 19(e)(2)). Data Privacy Act in DepEd Schools: RA 10173 for School Heads covers the rest of the Act as it applies to schools.

Anonymous, coded or confidential: what each means

Three words are used loosely in consent forms and research reports, and they promise different things.

Term What it means Who can tell whose row it is Under the Data Privacy Act
Anonymous No identity was ever collected: an unsigned questionnaire, a form with email collection off No one, unless a small group gives a respondent away Outside the Act if no one can be identified
Anonymized Identifiers removed and the remaining columns widened until no one can single out a learner; no key kept No one: identifying anyone is impossible or extremely impractical Outside the Act only if “truly anonymized” (NPC Advisory Opinion No. 2021-001)
Coded (pseudonymized) Names and LRNs replaced with codes; a key matches codes to names Whoever holds the key Still personal data (NPC Advisory Opinions No. 2018-029 and No. 2026-007)
Confidential The researcher knows who said what and keeps it from everyone else The researcher Personal data, protected by the researcher’s safeguards

The NPC’s test for “truly anonymized” is strict: the anonymization “must be irreversible, and done in such a way that it is impossible (or extremely impractical) to identify a data subject. There must be no way for the POEA or any other person to single out an individual in a given data set, from connecting two records within a data set (or between two separate data sets) and from any information in such data set” (Advisory Opinion No. 2021-001). An August 2026 opinion on corporate records puts it bluntly: “Replacing names with initials or codes is only pseudonymization when the person remains identifiable from position, compensation, or other context” (Advisory Opinion No. 2026-005).

Most action research keeps a key, because the pretest and posttest must be matched learner by learner. Such a study can promise confidentiality and coding, not anonymity. A consent letter can say so plainly:

Your child’s answers will be recorded under a code, never a name. Only the teacher-researcher will keep the list that matches codes to names, in a locked file at the school. No name, LRN or birth date will appear in the report or in the data file given to the Division Office.

Anonymity is the right promise only for a study that never keeps a key, such as a one-time survey with no names, and then only if its small groups do not give anyone away.

Which columns identify a learner

A class list holds three kinds of columns. Direct identifiers name a learner by themselves. Quasi-identifiers, which the National Institute of Standards and Technology (NIST) defines as identifiers “that by themselves do not identify a specific individual but can be aggregated and ‘linked’ with other information to identify data subjects” (NIST IR 8053, Sec. 3.4), point to a learner in combination. The study’s own measures are what a re-identified row exposes.

Column Kind What to do
Name, LRN, address, parent’s or guardian’s name and number, email, photo Direct identifier Delete, or move to the key with the code
Birth date Quasi-identifier that comes close to naming a learner Turn into an age or an age band, then delete
Sex, age, grade, section, sitio or barangay Quasi-identifier Keep only what the analysis uses; check the groups
Mother tongue, Indigenous Peoples group, religion, disability, 4Ps membership Quasi-identifier; IP group, religion and disability are sensitive personal information (RA 10173, Sec. 3(l)) Keep only when the research question needs it; a rare value often stands alone
Scores, answers, attendance, reading level What the study measures Keep; this is what a re-identified row reveals
Free-text answers and remarks May name people or events Read each one; replace names and details with general words

The LRN deserves its own word. It is a “permanent twelve (12)-digit number which the pupil, student or learner shall keep while completing the basic education program, regardless of transfer to another school” (DO 22, s. 2012, item 2), and “The LRN shall be incorporated in all documents, forms, examinations, surveys and databases which refer to a pupil, student or learner” (item 3). That makes it a key to every other record about the learner. NIST advises treating such numbers as direct identifiers “even though additional information is required to link them to an identity, because these forms of identification are extensively used and thus available for linking to identities” (NIST IR 8053, Sec. 3.1). For the same reason, a code must not be built from the LRN or the learner’s initials: the United States education department’s privacy office says a record code “may not be based on a student’s social security number or other personal information” (PTAC, Data De-identification: An Overview of Basic Terms, p. 5).

The report’s own words are a column too. An action research paper usually names its school, grade and section, so “the only 13-year-old girl” in the data set is the only 13-year-old girl in a section anyone can find.

How to anonymize a class list in Excel, step by step

The figure follows six rows of an invented multigrade class of 18 learners from the class list to the data set. Pick a step to see what it leaves on the sheet.

Six of 18 rows · every name and number invented

From class list to data set

Pick a step. The sheet shows what that step leaves; a column it takes away is struck through until the next step. Nothing you choose is kept or sent anywhere.

Step

Step 1 of 8Start from a copy of the class list

Copy the names, LRNs and other columns from the School Form 1 and the scores from the class record into a new workbook. Work on the copy; the originals stay where they are.

Everything below happens to the copy. The original forms remain the school’s records.

School Form 1 columns: DepEd Order No. 4, s. 2014, data element descriptions

Step 2 of 8Delete the columns the study never uses

Delete every column the research questions do not need: here the row number, mother tongue, religion, sitio and the parent’s number. Delete them, do not hide them.

A column that is not in the file cannot leak. The Data Privacy Act asks that personal information be “adequate and not excessive in relation to the purposes for which they are collected and processed.”

RA 10173, Sec. 11(d)

Step 3 of 8Shuffle the rows, give codes, and cut off the key

Add a helper column with =RAND(), fill it down and sort the table by it. Then give codes down the new order, paste them as values, and move the names and LRNs, with the codes, to a separate workbook: the key.

="L"&TEXT(ROW()-1,"00")In row 2 it gives L01, in row 3 L02, and so on.

A class list comes in an order, here boys first and then girls, each by surname. Codes given in that order would say that L01 is a boy whose surname starts early in the alphabet. A random sort breaks that link; Sweeney’s fix for leaks through row order is “randomly sorting the tuples.”

Sweeney 2002, Sec. 4.1

Step 4 of 8Turn birth dates into ages

Put the day the ages count to in one cell (here 08/17/2026, the pretest), compute each age, paste the ages as values, and delete the birth dates and the helper column.

=DATEDIF(D2,$K$1,"Y")D2 holds the birth date; K1 holds the day the ages count to.

A birth date nearly names a person by itself: with sex and a five-digit ZIP code, it “likely made” 87% of the 1990 United States population unique. An age in years is shared with classmates.

Sweeney 2000, Abstract

Step 5 of 8Count how many learners share each row

Beside the data, count the learners in the whole class with the same sex, age and grade as each row.

=COUNTIFS(B$2:B$19,B2,C$2:C$19,C2,D$2:D$19,D2)B is sex, C is age and D is grade, over the 18 rows of the class.

A 1 means no other learner fits the row, so anyone who knows the class knows whose scores it holds. L03, the only 12-year-old boy in Grade 5, and L10, the only 13-year-old girl in Grade 6, stand alone; so do two rows not shown here.

k-anonymity: Sweeney 2002, Definition 3

Step 6 of 8Widen the ages into bands, and count again

Replace each age with a two-year band, paste as values, and count again with the band in place of the age.

=IF(C2<=11,"10-11","12-13")The class’s ages run from 10 to 13.

The band puts L10 with two other girls of 12 or 13 in Grade 6. L03 still stands alone: no other Grade 5 learner is 12 or 13. So does one row not shown here: L09, the only girl of 10 or 11 in Grade 6.

Generalization: NIST IR 8053, Sec. 3.4

Step 7 of 8Leave out what still points to one learner

If the analysis never uses age, delete the band column and give the class’s age range in the paper instead (“aged 10 to 13”).

With sex and grade only, every row fits at least four learners. Removing a column is what NIST calls suppression; widening it, generalization.

Suppression: NIST IR 8053, Sec. 3.4

Step 8 of 8Save the data set and the key apart

Delete the count column, add a codebook sheet (each column, what it means, its codes), and save the data set as a new workbook. Keep the key in the school, where only the researcher can open it.

The data set is what the Research Management Guidelines ask for. A code is undone by whoever has the key, so the key never travels with the data: the lookup table “must be highly protected.”

DO 16, s. 2017, Sec. V.B.vii; NIST SP 800-188, Sec. 4.3.1

Key.xlsxStays in the school, opened only by the researcher

CodeLRNName
L03999999190089Dela Cruz, Jomar
L10999999190112Castillo, Andrea
L13999999210032Villanueva, Paolo
L14999999190145Torres, Lyka
L15999999200063Mendoza, Rica
L16999999200147Bautista, Carlo

Class list (copy).xlsxWorking copy.xlsxData set.xlsxStill names every learnerCodes in place of names; the key is cut offWhat may leave the school, with its codebook

No.deletedto the keynewCodedeletedto the keynewShuffledeletedto the keynewLRNdeletedto the keynewNamedeletedto the keynewSexdeletedto the keynewBirth datedeletedto the keynewAgedeletedto the keynewAge banddeletedto the keynewMother tonguedeletedto the keynewReligiondeletedto the keynewSitiodeletedto the keynewParent’s numberdeletedto the keynewGradedeletedto the keynewPretestdeletedto the keynewPosttestdeletedto the keynewSame sex, age, gradedeletedto the keynewSame sex, band, gradedeletedto the keynewSame sex, gradedeletedto the keynew
1L160.8813999999200147Bautista, CarloM02/09/20151110-11CebuanoCatholicCentro09•• ••• 44716813225
2L030.1172999999190089Dela Cruz, JomarM05/21/20141212-13CebuanoCatholicIlaya09•• ••• 09385611114
3L130.6081999999210032Villanueva, PaoloM01/30/20161010-11TagalogChristianIlaya09•• ••• 66125712234
4L100.4630999999190112Castillo, AndreaF07/03/20131312-13CebuanoCatholicIlaya09•• ••• 20546914134
5L150.7724999999200063Mendoza, RicaF11/12/20141110-11CebuanoCatholicCentro09•• ••• 739051016255
6L140.6597999999190145Torres, LykaF03/26/20141212-13CebuanoIglesia ni CristoIbabao09•• ••• 118761215234
2L030.1172999999190089Dela Cruz, JomarM05/21/20141212-13CebuanoCatholicIlaya09•• ••• 09385611114
4L100.4630999999190112Castillo, AndreaF07/03/20131312-13CebuanoCatholicIlaya09•• ••• 20546914134
3L130.6081999999210032Villanueva, PaoloM01/30/20161010-11TagalogChristianIlaya09•• ••• 66125712234
6L140.6597999999190145Torres, LykaF03/26/20141212-13CebuanoIglesia ni CristoIbabao09•• ••• 118761215234
5L150.7724999999200063Mendoza, RicaF11/12/20141110-11CebuanoCatholicCentro09•• ••• 739051016255
1L160.8813999999200147Bautista, CarloM02/09/20151110-11CebuanoCatholicCentro09•• ••• 44716813225

… and 12 more rows. Ages are counted to 08/17/2026, the day of the pretest.

An invented multigrade class. The formulas are Excel’s and work in Google Sheets; the column letters assume the sheet as drawn at that step, with headings in row 1.

The steps, with what the figure leaves out:

  1. Work on a copy. Copy what the study needs from the School Form 1 and the class record into a new workbook; the forms stay where they are.
  2. Delete what the study does not use. If the research questions never mention religion, mother tongue or address, those columns go. The Act asks that personal information be “adequate and not excessive in relation to the purposes for which they are collected and processed” (RA 10173, Sec. 11(d)). Delete a column (right-click its letter, Delete); hiding it keeps it in the file.
  3. Shuffle, then give codes. Add a column with =RAND(), fill it down, and sort the whole table by that column (Data, Sort). Type the codes down the sorted rows, or use ="L"&TEXT(ROW()-1,"00"), then copy the codes and paste them back as values (Paste Special, Values). A code made with ROW() belongs to the row’s place, not the learner, so it moves the moment the rows are sorted again, and RAND returns a new number “every time the worksheet is calculated.” Paste the codes as values before anything else. In Excel for Microsoft 365, Excel 2021 and Excel 2024, =SORTBY(A2:G19,RANDARRAY(ROWS(A2:G19))) gives a shuffled copy in one formula; paste it as values before giving codes, or it reshuffles at every edit. Excel 2016 and 2019 have neither function, so use the helper column there.
  4. Cut off the key. Copy the Code column, with the Name and LRN columns, into a separate workbook, then delete Name and LRN from the data set. That workbook is the only link back to the learners, so it never travels with the data set. NIST’s guide for government data sets says “the lookup table or the information for the transformation must be highly protected” and that “the pseudonym must be randomly assigned” (NIST SP 800-188, Sec. 4.3.1).
  5. Turn birth dates into ages. Type the day ages are counted to in one cell (the day of the pretest, for example) and put =DATEDIF(D2,$K$1,"Y") in a new column beside the birth dates, where D2 holds the birth date and K1 the day. Paste the ages as values, then delete the birth dates and the RAND column.
  6. Count who matches each row. In a spare column, =COUNTIFS(B$2:B$19,B2,C$2:C$19,C2,D$2:D$19,D2) counts the learners with the same sex (B), age (C) and grade (D) as the row. A 1 means the row fits one learner only.
  7. Widen or leave out what stands alone. Replace exact ages with bands such as =IF(C2<=11,"10-11","12-13") for a class aged 10 to 13 (or =FLOOR(C2,2)&"-"&(FLOOR(C2,2)+1) for any ages) and count again. If a row still stands alone and the analysis does not use the column, delete it and give the range in the report instead (“aged 10 to 13”). NIST calls the first move generalization and the second suppression (NIST IR 8053, Sec. 3.4).
  8. Save the data set apart. Paste everything as values, delete the helper columns, add a codebook sheet (each column, what it means, its codes), and save the data set as a new workbook. Then inspect the file for what Excel keeps out of sight (see below).

A study that matches a pretest and a posttest gives codes before the first test is scored, writes each code on the learner’s answer sheet, and encodes scores by code, as Coding Guide and Data Cleaning in Excel for Action Research sets out. Names then never enter the scoring file at all. Google Sheets has RAND, SORT, RANDARRAY, DATEDIF and COUNTIFS, so the same steps work there.

Why shuffle before giving codes

Codes given in the class list’s order pass that order on. School forms printed from the Learner Information System keep the system’s order of names, which may not be edited (DepEd Order No. 11, s. 2018), and DepEd’s own School Form 1 template for senior high lists male learners in a block above female learners. Codes given down such a list would tell a reader that a low code is a boy. The Action Research Toolkit’s coding guide has the same flaw: papers “can be arranged in some kind of sequence, say alphabetically, then assigned an ID number” (Sec. 5.1). Latanya Sweeney, who defined k-anonymity, names a related leak through row order, the “unsorted matching attack,” and gives the cure: “randomly sorting the tuples” (Sweeney, 2002, Sec. 4.1).

How age and sex can still point to one learner

The cards below are the 18 rows of the invented class’s data set, already coded. Choose which columns the data set keeps; each card counts the learners who share its description.

An invented class of 18 · coded L01 to L18 · no names, no LRNs

Which rows point to one learner?

Choose the columns the data set keeps. Each card is one row, and its number counts the learners in the class who share that description. Nothing you choose is kept or sent anywhere.

Columns the data set keeps
Age
  1. L01Female10 years old10–11 years oldGrade 5Sitio CentroCode onlyOnly this learnerlearners fit
  2. L02Female12 years old12–13 years oldGrade 6Sitio CentroCode onlyOnly this learnerlearners fit
  3. L03Male12 years old12–13 years oldGrade 5Sitio IlayaCode onlyOnly this learnerlearners fit
  4. L04Female10 years old10–11 years oldGrade 5Sitio IlayaCode onlyOnly this learnerlearners fit
  5. L05Male11 years old10–11 years oldGrade 6Sitio IbabaoCode onlyOnly this learnerlearners fit
  6. L06Male11 years old10–11 years oldGrade 5Sitio CentroCode onlyOnly this learnerlearners fit
  7. L07Male12 years old12–13 years oldGrade 6Sitio IlayaCode onlyOnly this learnerlearners fit
  8. L08Female11 years old10–11 years oldGrade 5Sitio IbabaoCode onlyOnly this learnerlearners fit
  9. L09Female11 years old10–11 years oldGrade 6Sitio CentroCode onlyOnly this learnerlearners fit
  10. L10Female13 years old12–13 years oldGrade 6Sitio IlayaCode onlyOnly this learnerlearners fit
  11. L11Female10 years old10–11 years oldGrade 5Sitio CentroCode onlyOnly this learnerlearners fit
  12. L12Male12 years old12–13 years oldGrade 6Sitio CentroCode onlyOnly this learnerlearners fit
  13. L13Male10 years old10–11 years oldGrade 5Sitio IlayaCode onlyOnly this learnerlearners fit
  14. L14Female12 years old12–13 years oldGrade 6Sitio IbabaoCode onlyOnly this learnerlearners fit
  15. L15Female11 years old10–11 years oldGrade 5Sitio CentroCode onlyOnly this learnerlearners fit
  16. L16Male11 years old10–11 years oldGrade 6Sitio CentroCode onlyOnly this learnerlearners fit
  17. L17Male10 years old10–11 years oldGrade 5Sitio IbabaoCode onlyOnly this learnerlearners fit
  18. L18Male12 years old12–13 years oldGrade 6Sitio CentroCode onlyOnly this learnerlearners fit

Sex, age in years, grade and sitio kept: 14 of the 18 rows fit only one learner each (k = 1).

Sex, age in years and grade kept: 4 of the 18 rows fit only one learner each (k = 1).

Sex, age in years and sitio kept: 8 of the 18 rows fit only one learner each (k = 1).

Sex and age in years kept: 1 of the 18 rows fits only one learner (k = 1).

Sex, a two-year age band, grade and sitio kept: 13 of the 18 rows fit only one learner each (k = 1).

Sex, a two-year age band and grade kept: 2 of the 18 rows fit only one learner each (k = 1).

Sex, a two-year age band and sitio kept: 6 of the 18 rows fit only one learner each (k = 1).

Sex and a two-year age band kept: every row fits at least 3 learners (k = 3).

Sex, grade and sitio kept: 8 of the 18 rows fit only one learner each (k = 1).

Sex and grade kept: every row fits at least 4 learners (k = 4).

Sex and sitio kept: every row fits at least 2 learners (k = 2).

Sex kept: every row fits at least 9 learners (k = 9).

Age in years, grade and sitio kept: 7 of the 18 rows fit only one learner each (k = 1).

Age in years and grade kept: 2 of the 18 rows fit only one learner each (k = 1).

Age in years and sitio kept: 3 of the 18 rows fit only one learner each (k = 1).

Age in years kept: 1 of the 18 rows fits only one learner (k = 1).

A two-year age band, grade and sitio kept: 3 of the 18 rows fit only one learner each (k = 1).

A two-year age band and grade kept: 1 of the 18 rows fits only one learner (k = 1).

A two-year age band and sitio kept: 1 of the 18 rows fits only one learner (k = 1).

A two-year age band kept: every row fits at least 7 learners (k = 7).

Grade and sitio kept: every row fits at least 2 learners (k = 2).

Grade kept: every row fits at least 9 learners (k = 9).

Sitio kept: every row fits at least 4 learners (k = 4).

Only the codes kept: every row fits all 18 learners (k = 18).

An invented class. A row that fits only one learner tells anyone who knows the class whose scores it holds; the smallest number on the cards is the data set’s k.

A row with a count of 1 fits one learner. A classmate, a parent or a teacher who knows that the only 12-year-old boy in Grade 5 is in the study now knows his scores. Sweeney found the same at the scale of a country: “87% (216 million of 248 million) of the population in the United States had reported characteristics that likely made them unique based only on {5-digit ZIP, gender, date of birth}” (Sweeney, 2000, Abstract).

The smallest count on the cards is the data set’s k. Sweeney defined it this way: “A release provides k-anonymity protection if the information for each person contained in the release cannot be distinguished from at least k-1 individuals whose information also appears in the release” (Sweeney, 2002, Abstract). The COUNTIFS column of step 6 computes each row’s group, and its smallest value is k.

No Philippine rule sets a minimum k for a research data set. Statisticians’ practice gives a floor: “Many statisticians consider a cell size of 3 to be the absolute minimum needed to prevent disclosure, though larger minimums (e.g., 5 or 10) may be used to further mitigate disclosure risk” (PTAC, Frequently Asked Questions: Disclosure Avoidance, p. 3). The United Kingdom’s National Health Service sets k at five for strong anonymization of published health data (ICO, Anonymisation). In the invented class, k reaches 3 or more only when the data set keeps two of these columns at most: sex and grade (k = 4), or sex and the age band (k = 3). A large k does not protect a group whose members all share one result: if all five boys in Grade 6 scored below the passing mark, the data set tells everyone that each of them did.

The learners most at risk are the ones who differ: the overage learner, the only learner with a disability, the only Indigenous Peoples learner, the transferee. The United Kingdom’s data protection regulator warns that “Identifiability risk can arise where one person or group knows a great deal about another person” (ICO, Anonymisation), and in a school that group is the class itself. A learner who stands alone on a column can be protected by widening it (“12 and older”), by leaving the column out, or, if the study needs it, by keeping the data set inside the school.

Small numbers in the report’s tables

The same arithmetic applies to the tables in the paper. If a table says that all four girls in Grade 6 scored below the passing mark, it tells every reader each girl’s result. When one of two groups is hidden and the other group and the total are shown, “the suppressed data can be fully recovered” by subtraction (NCES, SLDS Technical Brief 3, p. 9). Report a group of fewer than three learners as “fewer than 3”, and hide one more cell in the same row or column so the hidden one cannot be worked out.

Hiding a column is not anonymizing

A workbook carries more than its visible sheet. Turn on the X-ray to see what a recipient can still find in a file that looks clean, and pick a layer for what it holds and how to remove it.

An invented workbook

What the workbook still holds

On screen the file shows only coded rows. Turn on the X-ray to see the layers under the sheet, or pick one. Nothing you choose is kept or sent anywhere.

Layers under the sheet

A hidden column

What it holds
The Name column, hidden between Code and Sex. A double line between the column letters A and C, and the missing B, are the only signs of it.
How a recipient finds it
Double-click the double line between the column letters, or select columns A to C, right-click and choose Unhide.
How to remove it
Delete the column instead of hiding it. The Document Inspector finds hidden rows and columns that hold data and can remove them.

Microsoft Support, Remove hidden data and personal information by inspecting documents

A hidden sheet

What it holds
The sheet the data set was made from, with every name, LRN and birth date, hidden behind the Data set tab.
How a recipient finds it
Right-click any sheet tab and choose Unhide.
How to remove it
Delete the sheet. The Document Inspector lists hidden worksheets and can delete them.

Microsoft Support, Remove hidden data and personal information by inspecting documents

A PivotTable’s saved rows

What it holds
A summary table made from the class list. A PivotTable keeps its own copy of the source data in the file, so the names stay in the workbook after the class list sheet is deleted.
How a recipient finds it
Double-click a number in the PivotTable: Excel puts the detail data behind it on a new worksheet.
How to remove it
Make summaries from the coded data only, or copy the PivotTable and paste it as values into the data set. The Document Inspector flags a PivotTable’s saved data but cannot remove it, and Microsoft says the option “Save source data with file” “should not be used to manage data privacy.”

Microsoft Support, PivotTable options; Expand, collapse, or show details in a PivotTable or PivotChart; Remove hidden data and personal information by inspecting documents

A comment

What it holds
A comment left while encoding. It names no one, but it ties a health condition to L03, and anyone who saw who missed Monday’s test knows who L03 is.
How a recipient finds it
Point at the small marker in the cell’s corner, or open the comments from the Review tab.
How to remove it
Delete it, or let the Document Inspector find comments and remove them all.

Microsoft Support, Remove hidden data and personal information by inspecting documents

A link to the key

What it holds
A lookup formula that once fetched codes from the key. The workbook saves the linked file’s path and the names of its worksheets, which tell the recipient that a key exists and where it was kept.
How a recipient finds it
The formula bar shows the link in each cell, and Excel lists linked workbooks under the Data tab.
How to remove it
Paste the column as values, so no formula points to the key. The Document Inspector flags external links but cannot remove them.

Microsoft Support, Remove hidden data and personal information by inspecting documents

The file’s properties

What it holds
The author field, and the name of whoever last saved the file, which Excel keeps on its own. They name the teacher, and so the school and the class.
How a recipient finds it
File, Info, then the properties listed beside the file’s details.
How to remove it
The Document Inspector removes document properties and personal information, including the user name.

Microsoft Support, Remove hidden data and personal information by inspecting documents

Earlier versions

What it holds
On OneDrive, SharePoint or Google Drive, earlier versions of the working copy, including the one from before the names were deleted.
How a recipient finds it
Anyone with permission to edit a shared Google Sheet can browse its earlier versions; in Excel, Version history opens from the file’s title.
How to remove it
Save the data set as a new file and send that file, not a share link to the working copy.

Microsoft Support, View previous versions of Office files; Google Docs Editors Help, Find what’s changed in a file

The file name

What it holds
The section’s name in the file name tells the recipient which class the rows belong to.
How a recipient finds it
It travels with the file.
How to remove it
Save the data set under a plain name, such as reading-study-data-set.xlsx.

An invented workbook. Where each layer is found and removed follows Microsoft’s and Google’s help pages for the current versions of Excel and Google Sheets; menu names differ a little between versions.

Excel’s Document Inspector finds most of these layers: File, Info, Check for Issues, Inspect Document. It finds comments, document properties and personal information, headers and footers, hidden rows and columns that hold data, hidden worksheets and invisible objects, and removes each kind with Remove All. It also flags PivotTables, external links and other items that may keep “cached data,” but “can’t remove these items for you because it might cause the workbook to not work properly,” and it cannot touch the earlier versions a shared drive keeps. Microsoft advises running it on a copy, “because it is not always possible to restore the data that the Document Inspector removes.” Protecting a sheet with a password does not help either: “Worksheet level protection isn’t intended as a security feature” (Microsoft Support, Protect a worksheet). Save the cleaned data set as a new file with a plain name, and send that file, not a link to the working copy.

Anonymous surveys and interview transcripts

A Google Forms survey is anonymous only if it collects nothing that names the respondent. Do not set “Collect email addresses” to Verified or Responder input, and leave out questions that ask for a name, an LRN or a section that holds few learners. “Limit to 1 response” makes respondents sign in to Google, though “Their usernames won’t be recorded unless you turn on the ‘Responses’ setting to collect email addresses,” and a file-upload question also requires sign-in. Turn off “View results summary” (Settings, Presentation) on a form about learners: the summary shows “full text responses or charts for each question” to “anyone who can respond to the form” (Google Docs Editors Help, Publish & share your form with responders). Survey Research With Google Forms for Action Research covers the other settings.

Small groups give respondents away in surveys of teachers too. If a school has one male Grade 3 teacher, a survey that asks for sex and grade taught names him. Ask for what the analysis needs, in groups large enough to hide one person.

Interview and focus group transcripts need the same care in words. Replace each name with a tag used the same way in every transcript (“[Teacher 1]”, “[the barangay captain]”), and widen details that point to one person (“the only teacher who transferred from Manila” becomes “a teacher who transferred”). The UK Data Service warns that “Removing names alone is rarely sufficient” and that automatic search-and-replace can miss variations of a name. Interview and FGD Guide Questions for Action Research covers recording and transcribing.

Before pasting data into an AI tool

DepEd’s AI guidelines, DepEd Order No. 003, s. 2026, limit AI in research to “ancillary support functions, such as the analysis of processed data, grammatical refinement, and citation verification,” and add: “the input, processing, or storage of personal and sensitive information within AI systems is categorically prohibited” (DO 003, s. 2026, Sec. VI.1.3). Documents “that contain personally identifiable information shall not be uploaded, shared, or processed through AI platforms” (Sec. VII.1.1). The same item allows one exception: such data may be processed “within internally governed DepEd AI systems … solely for approved research or operational purposes, subject to data minimization, anonymization, consent where applicable, and appropriate privacy and security safeguards.” A public chatbot is not such a system. Since the NPC treats coded data as personal data, a coded class list is not cleared for an AI tool by its codes alone; only a table that no longer describes individual learners, such as the class’s mean scores, is safely outside the ban. DO 003 does not draw that line in so many words. AI in DepEd Action Research covers the declaration DepEd asks for.

What this means for school heads

  • Keep the key in the school. The key file belongs on a school computer or a school account that only the researcher can open, not on a personal phone or a shared drive. Access to learner data is on a “need-to-know basis” (DO 017, s. 2025, Sec. X.B), and files carried on a USB flash drive must be encrypted when that transfer cannot be avoided (NPC Circular No. 2023-06, Sec. 25).
  • Open the data set before endorsing a completed study. Look for names, LRNs and birth dates; run the Document Inspector; check that no PivotTable’s source holds names; and scan the codebook for columns no research question uses.
  • Decide when the key is deleted. Personal data are kept “only for as long as necessary for the fulfillment of the purposes for which the data was obtained” (RA 10173, Sec. 11(e)) and “shall not be retained in perpetuity in contemplation of a possible future use yet to be determined” (IRR, Sec. 19(e)(3)). The Action Research Toolkit lets documents that bear names be shredded “after one year of research dissemination” (Sec. 2.1, Protocol 3). Treat the key the same way: once the study has been disseminated and no follow-up needs the codes, delete it, and the data set comes close to anonymized.
  • Give outside researchers only what the approved study needs. A graduate student studying the school’s learners needs the coded columns of the approved study, not School Form 1. Coded data are still personal data, so DO 017’s rule against disclosing learners’ personal information to third parties without consent applies to them.
  • Teach the steps at a Learning Action Cell. Coding, the key and the count can be practiced on an invented class list like the one in the figures.

These tasks fall under PPSSH indicator 2.1.2, “Manage school data and information using technology, including ICT, to ensure efficient and effective school operations” (PPSSH Domain 2: Managing School Operations and Resources), and support indicator 1.4.2, on using research findings to improve school performance (PPSSH Domain 1: Leading Strategically). School Research Committee in DepEd covers the committee that reviews a school’s studies.

Frequently asked questions

Can you anonymize data in Excel?

Yes, though Excel has no single command for it. Anonymizing in Excel means deleting identifying columns, replacing names with codes after a random shuffle, moving the key to another file, widening ages into bands, and counting matches with COUNTIFS until no row fits one learner (NIST IR 8053, Sec. 3.4). Run the Document Inspector before sharing the file (Microsoft Support).

Is coded data the same as anonymous data?

No. As long as a key matches codes to names, the data are coded, or pseudonymized, and the NPC treats them as personal data (Advisory Opinion No. 2026-007). Data become anonymous only when identifying a learner from them is “impossible (or extremely impractical)” (Advisory Opinion No. 2021-001).

Does DepEd require removing names from the executive data set?

DO 16, s. 2017 asks for the data set in Excel and says nothing about names (Sec. V.B.vii). The duty comes from elsewhere: DO 22, s. 2012 keeps learners’ identities confidential (item 9), the Quality Control Checklist DepEd uses to accept BERF-funded research and archive studies in E-Saliksik requires completed research to comply with the Data Privacy Act (DM 028, s. 2022), and the Act’s research exemption reaches only as far as the research needs.

What is k-anonymity?

A data set is k-anonymous when every row’s description fits at least k learners in the data set, so no one can be told apart from at least k−1 others (Sweeney, 2002). In Excel, a COUNTIFS over the descriptive columns gives each row’s count, and the smallest count is k.

Are Google Forms surveys anonymous?

Only when the form collects no email address and no question asks for a name, an LRN or a small group that gives a respondent away. A form set to “Limit to 1 response” makes respondents sign in to Google, though it does not record their usernames unless email collection is on (Google Docs Editors Help, Publish & share your form with responders).

Sources

DepEd issuances

  • DepEd Order No. 16, s. 2017, Research Management Guidelines, March 20, 2017: Secs. III, V.B.vii, V.B.ix and VI.C; Annex 6; the BERF deliverables tables.

  • DepEd Order No. 026, s. 2021, Amendment to DepEd Order No. 016, s. 2017 (Research Management Guidelines), June 21, 2021: item 5.

  • DepEd Memorandum No. 028, s. 2022, Adopting Tools to Improve Quality Management of Completed Research at the Department of Education, April 5, 2022: Enclosure III.b; instruction guide, action research items 10 and 11.

  • DepEd Order No. 014, s. 2022, Adoption of E-Saliksik: The DepEd Research Portal, April 4, 2022: Sec. VI.b.

  • DepEd Order No. 22, s. 2012, Adoption of the Unique Learner Reference Number, March 20, 2012: items 2, 3 and 9.

  • DepEd Order No. 017, s. 2025, Revised Basic Education Enrollment Policy, June 13, 2025: Secs. X.B and X.C.

  • DepEd Order No. 003, s. 2026, Foundational Guidelines on Artificial Intelligence (AI) in Basic Education, February 20, 2026: Secs. VI.1.3 and VII.1.1.

  • Department of Education and the Basic Education Sector Transformation (BEST) Program, Action Research Toolkit: Teacher’s Guide, 2019: Secs. 2.1 (Protocol 3), 4.1, 4.2 and 5.1.

  • DepEd Region IV-A CALABARZON, Regional Memorandum No. 158, s. 2023, Submission of Data and Documents on Utilization of BERF-Funded Research Findings, March 17, 2023: items 1 and 2, as forwarded by the Schools Division of Tayabas City (Division Memorandum No. 160, s. 2023).

  • Schools Division of Tayabas City, Division Memorandum No. 137, s. 2021, Reconstitution of the Schools Division Research Committee (SDRC) and Specifying Roles Therein, April 12, 2021: enclosed Division Research Management Guidelines, Sec. IV.F (p. 8).

  • Schools Division of Bayugan City, Division Memorandum No. 348, s. 2025, Reiteration on the Submission of Copies of Completed Research Conducted in Schools and the Schools Division Office of Bayugan City Involving DepEd Personnel and Learners for Calendar Year 2025, October 28, 2025: items 4(b) and 6.

  • DepEd Order No. 11, s. 2018, Guidelines on the Preparation and Checking of School Forms, March 7, 2018.

Law and National Privacy Commission issuances

Research ethics and methods

This guide explains DepEd rules and the Data Privacy Act for study and professional awareness. It is not legal advice; for a specific case, consult your Schools Division Office’s legal unit or its data protection officer.

Written by Mark Anthony Llego for NASH Reviewer. Sources: DepEd Orders No. 16, s. 2017, No. 22, s. 2012, No. 017, s. 2025 and No. 003, s. 2026; DepEd Memorandum No. 028, s. 2022; RA 10173 and its IRR; NPC issuances and advisory opinions; NEGRIHP 2022; Sweeney (2000, 2002); NIST; PTAC; NCES; ICO; UK Data Service.